Development of an object-oriented infrared imaging system simulator and its application to multi-spectral infrared imaging - CaltechTHESIS
CaltechTHESIS
A Caltech Library Service
About
Browse
Deposit an Item
Instructions for Students
Development of an object-oriented infrared imaging system simulator and its application to multi-spectral infrared imaging
Citation
Springfield, Christopher D.
(1998)
Development of an object-oriented infrared imaging system simulator and its application to multi-spectral infrared imaging.
Dissertation (Ph.D.), California Institute of Technology.
doi:10.7907/jcn5-z151.
Abstract
This thesis describes research efforts undertaken to investigate passive, multi-spectral infrared imaging. Although many applications of multi-spectral infrared imaging may exist, the high cost of developing viable multi-spectral technologies has limited research into, and subsequent exploitation of, these applications. Our efforts attempt to find a way to minimize the cost of this research while concurrently investigating one of the possible applications of multi-spectral infrared imaging using current infrared imaging technology. The outcome is an object-oriented, infrared imaging system simulator, called IRIMAGE, and a series of experiments and simulations that confirm the viability of gaseous pollution detection using passive, multi-spectral infrared imaging.
IRIMAGE is a flexible tool capable of applications research and basic infrared system design. This combination makes it a cost effective tool for researching the applications of multi-spectral IR imaging and the technological requirements they require. We present the physical and computational concepts that underly the simulation as well as certain computational advances made during IRIMAGE's development. A comprehensive discussion of the primary objects that make up IRIMAGE and how the simulation works is also provided. Since the reliability of a simulation depends on experimental verification of its output, we also present the results of this verification.
Besides verifying IRIMAGE, these experiments investigated detecting gaseous pollutants using passive, multi-spectral IR imaging. The thesis describes the imaging system we used and the theoretical background of these experiments. For each experiment, we describe the experimental setup and how IRIMAGE simulated the experiment. Finally, we compare the experimental and simulation results. Although these experiments verify IRIMAGE and demonstrate how gaseous pollutants can be detected using passive, multi-spectral IR imaging, further research is necessary and certain technological advances must be made before this application can be exploited. More information about IRIMAGE is available on the web at www.ssdp.caltech.edu.
Item Type:
Thesis (Dissertation (Ph.D.))
Degree Grantor:
California Institute of Technology
Division:
Engineering and Applied Science
Major Option:
Applied Physics
Thesis Availability:
Public (worldwide access)
Research Advisor(s):
McGill, Thomas C.
Thesis Committee:
Unknown, Unknown
Defense Date:
11 November 1997
Record Number:
CaltechETD:etd-01312008-150151
Persistent URL:
DOI:
10.7907/jcn5-z151
Default Usage Policy:
No commercial reproduction, distribution, display or performance rights in this work are provided.
ID Code:
422
Collection:
CaltechTHESIS
Deposited By:
Imported from ETD-db
Deposited On:
20 Feb 2008
Last Modified:
19 Apr 2021 22:36
Thesis Files
Preview
PDF (Springfield_cd_1998.pdf)
- Final Version
See Usage Policy.
24MB
Repository Staff Only:
item control page
CaltechTHESIS is powered by
EPrints 3.3
which is developed by the
School of Electronics and Computer Science
at the University of Southampton.
More information and software credits
DEVELOPMENT OF AN OBJECT-ORIENTED
INFRARED IMAGING SYSTEM SIMULATOR
AND ITS APPLICATION TO
MULTI-SPECTRAL INFRARED IMAGING
Thesis by
Christopher D. Springfield
In Partial Fulfillment of the Requirements
for the Degree of
Doctor of Philosophy
California Institute of Technology
Pasadena, California
1998
(Submitted November 11, 1997)
il
Christopher D. Springfield
ill
Acknowledgements
To say that I couldn’t have done this alone would be a universal understatement. My
experiences at Caltech have been extraordinary and I leave here with much more than
I came with. At the center of my Caltech life was the McGill Group. If it were not for
Dr. Tom McGill’s leadership, inspiration, and love of lots of cool toys, much of what
I accomplished here would not have been possible. I owe him a debt of gratitude for
being patient and giving me the opportunities to expand both my scientific and my
creative sides.
One of Tom’s many skills is his gift of finding the right people for the right job.
Dr. Gerry Picus was the right person in my case. With Gerry’s help and guidance,
I was able to wrangle this project in and define a clear direction for it to take. At
the same time, he showed true interest in the other activities of my life and always
seemed eager for a good discussion of how to solve the world’s problems. I will always
consider him a friend.
Throughout my years in the McGill group, the faces have changed, but the per-
sonality of the group remains the same. I have never worked with such a group of
broad-minded people. Every single person seems able to pursue brilliant scientific
endeavors without losing sight of the goal in life, which is to live. Among those I have
had the honor to work with, I wish to thank Wesley Salzillo for helping me get the
camera running and finishing those pesky experiments; David Ting for providing that
always calm but excited outlook on life; Harold Levy for being the doer of things yet
to be done; Rob Miles, Johannes Swenberg, Per-Olav Petterson and Ron Marquardt
for expanding my moralistic, economic and political horizons; Mike Wang and Yixin
Liu for being patient enough to wait until this thesis was finished to get their wedding
videos; and Marcia Hudson who was always there to fix one administrative headache
after another. To those still in the McGill group, I thank you for continuing the
group’s glorious traditions and bringing your own personalities to the mix. Long live
iv
the spirit that binds us.
Outside of my McGill group experience, I must thank those individuals who
touched my life and pushed me to finish what seemed to be an impossible task.
At the top of that list is Kent Bradford, roommate, friend, confidant. Were it not
for Kent, I doubt I would be the physicist I am today. He pushed and pulled me
through the early years and stood by as friend in the later ones. I am proud to call
him one of the best friends of my life. Among the rest, I must thank David Santiago,
Mike O’Neal, Richard and Carolyn Doherty, and Shirley Marnaeus. They inspired
me to pursue my dreams and were always there to lend me a hand when I needed it.
Finally, I must thank everyone at Digital Domain for giving me eight months of pure
pleasure and letting me grab hold of the brass ring for awhile. Never in my wildest
dreams would I think I would have the chance to work on a potential Academy Award
winning film and earn my Ph.D. in the same year. How glorious life can be.
Finally, there are four people I must thank above all the rest. They are ones that
made it happen. I must thank my parents, Joe and Alice Springfield. Were it not
for their love, support and encouragement, I would have stopped long ago. I must
also thank my late grandfather, Richard Smeltzer, who died long before I started this
journey. His memory was my inspiration, my drive. Everything I have accomplished
is my way of honoring such a wonderful person. Finally, I must thank my fiancée
Diana Lavely. As my love, my life, and my soulmate, she made sure this journey
came to a happy close. It is to her that I owe so much. I thank God that I have an
entire lifetime to make it up to her.
Abstract
This thesis describes research efforts undertaken to investigate passive, multi-spectral
infrared imaging. Although many applications of multi-spectral infrared imaging may
exist, the high cost of developing viable multi-spectral technologies has limited re-
search into, and subsequent exploitation of, these applications. Our efforts attempt to
find a way to minimize the cost of this research while concurrently investigating one
of the possible applications of multi-spectral infrared imaging using current infrared
imaging technology. The outcome is an object-oriented, infrared imaging system sim-
ulator, called IRIMAGE, and a series of experiments and simulations that confirm the
viability of gaseous pollution detection using passive, multi-spectral infrared imaging.
IRIMAGE is a flexible tool capable of applications research and basic infrared sys-
tem design. This combination makes it a cost effective tool for researching the appli-
cations of multi-spectral IR imaging and the technological requirements they require.
We present the physical and computational concepts that underly the simulation as
well as certain computational advances made during IRIMAGE’s development. A
comprehensive discussion of the primary objects that make up IRIMAGE and how
the simulation works is also provided. Since the reliability of a simulation depends on
experimental verification of its output, we also present the results of this verification.
Besides verifying IRIMAGE, these experiments investigated detecting gaseous pol-
lutants using passive, multi-spectral IR imaging. The thesis describes the imaging
system we used and the theoretical background of these experiments. For each ex-
periment, we describe the experimental setup and how IRIMAGE simulated the ex-
periment. Finally, we compare the experimental and simulation results. Although
these experiments verify IRIMAGE and demonstrate how gaseous pollutants can be
detected using passive, multi-spectral IR imaging, further research is necessary and
certain technological advances must be made before this application can be exploited.
More information about IRIMAGE is available on the web at www.ssdp.caltech.edu.
vi
Contents
Acknowledgements
Abstract
Glossary of Acronyms
Glossary of Terminology
1 Overview of Research
1.1 Motivation. 2... .
1.2 The IRIMAGE Simulation ..........020..20.0.0..0....0..
1.3 Multi-Spectral Infrared Imaging Experiments. .............
1.4 Outline of Thesis ... 2... 0.020200... 0000000000004
Bibliography
References... 0
2 IRIMAGE: An Infrared Imaging System Simulator
2.1
2.2
2.3
Introduction. . 2...
Physical Basis of IRIMAGE ...................0...
2.2.1 Blackbody Radiation ....................0..
2.2.2 Modeling the Atmosphere using MODTRAN..........
2.2.3 Determining Optical Path using Ray Tracing. .........
2.2.4 Computing the Incident Radiation upon the FPA .......
2.2.0 Modeling Detector Response. ...............00..
2.2.6 Modeling Background Fluctuation Noise ............
Computational Concepts... 2... ..0.0.0.00 000000000000.
2.3.1 The World Coordinate System (WCS) .............
ili
Xiv
Xvi
11
16
17
17
vil
2.3.2 The Versatile Grid Object (VGO) .. 2... ...0.0...0... 48
2.3.3 Interpolating 2D Image Data onto the VGO .........., 50
2.3.4 Interpolation of Volumetric Data onto the VGO ........ 54
2.3.5 The Ray Object... 2.2.0.0. 0.0 20.2.0. 0.022.000. o7
2.4 Major Elements of IRIMAGE ............0........4. 58
2.4.1 The Scene Object... ........0.0. 0.202.200 2200.4 58
2.4.2 The Atmosphere Object ..................0.. 60
2.4.3 The Focal Plane Array (FPA) Object... .....0..00202., 65
2.4.4 The Optics Object 2... 2.0.0.0. 0. 00000000058. 69
2.4.5 The Detector Object... ......... 0.0.2.0. .004. 73
2.4.6 The Output Object... 2... 0.020.002.2000 0004 75
2.5 The IRIMAGE Simulation... 2... ...0.0.0..0...0...00.. 76
2.5.1 The Graphical User Interface (GUI)... ......0.0.0002.. 76
2.5.2 How IRIMAGE Computes An Image .............. 77
2.5.3 Two Test Cases... 2. ee 80
2.6 Conclusions .. 2... 87
Bibliography 89
References... 2 0. 89
3 Multi-Spectral Experiments and the Verification of IRIMAGE 91
3.1 Introduction... 2... 2. 91
3.2 The Imaging System ...........0... 2.2.2.0 0000004 93
3.2.1 The Amber AE-256 Imaging System .............. 93
3.2.2 Simulating the Imaging System using IRIMAGE........ 97
3.2.3 Comparing Experimental Images with Simulated Images from
IRIMAGE. . 2... ee 98
3.3 Pollution Detection Using Multi-Spectral Imaging Methods..... . 104
3.3.1 The Vibration and Rotation Modes of a Molecule ...... . 105
3.3.2 Active and Passive Detection of Gaseous Pollutants ...... 111
vill
3.4 The Relative Temperature Experiments (RTE)............. 113
3.4.1 Experimental Setup... 2... ..0.0202.0.2 2.20.00 000. 114
3.4.2. Experimental Procedure ...........2...... 0048. 116
3.4.3 Simulation Setup .......... 02.000 000022 2 ae 118
3.4.4 Comparison of Image Results .................. 121
3.5 The Methane Experiments .............0....-2. 2.224. 127
3.0.1 Experimental Setup... .......0...0.2...22. 202004 128
3.5.2. Experimental Procedure ...................0.. 130
3.5.3 Simulation Setup .. 2... ..0.0.0.0.020.0 0.22000 0000. 131
3.5.4 Comparison of Results ................. 0.00.0. 134
3.6 The Gas Cell Experiments ..........0.......2..22004 141
3.6.1 Experimental Setup. ............0....-.. 2.00.4. 143
3.6.2. Experimental Procedure ..................0.0. 146
3.6.3 Simulation Setup ..............2.. 2.20.00 00.% 149
3.6.4 Comparison of Results ...........0.. 0.00.22 004 151
3.7 Conclusions .. 2... 162
Bibliography 165
References. 2 2. 0. 165
Appendix 167
A Block Diagrams of IRIMAGE 167
B Interpolating Volumetric Data onto a 3D Orthogonal Grid 173
B.1 Introduction... 2... 173
B.2 The Versatile Grid Object (VGO) .. 2... 2.00... ee, 175
B.2.1 The Grid Coordinate System (GCS) ...........002, 175
B.2.2 Data Storage inthe VGO .................... 179
B.3 Defining Bounded Volumes for Interpolating on the VGO....... 182
B.3.1 Defining Bounded Surfaces Using Polygonal Surface Modeling 182
B.3.2. The Universal Primitive Object (UPO) Model ......... 185
B.4 Interpolating a UPO ontoaVGO .........2..2..0....0... 193
B.4.1 Intercepting a Cutting Plane with the Surface of a UPO-based
Geometric Model... 2... 0.0.0.0... 002.022.0048. 194
B.4.2 Ordering the Intercepted Closed Curves ............ 197
B.4.3 Using Closed Curves to Store Dataon VGO ........2.., 198
B.4.4 The Slice and Dice Algorithm .................. 203
B.5 Application of the VGO and the UPO in the IRIMAGE Simulator . . 205
B.5.1 Generating an Atmospheric Profilein IRIMAGE........ 206
B.5.2 Example of Various UPOs Used inIRIMAGE ......... 209
B.6 Conclusions .. 20... ee 211
Bibliography 214
References . 2 0. 0 214
C Example of Parameter Files Used by IRIMAGE 216
C.1 The General Parameter File... ......0..0..........0.. 217
C.2 The Scene Object Parameter File .................... 218
C.3 The Atmosphere Object Parameter File... .............. 223
C.4 The FPA Object Parameter File...................0.. 231
C.5 The Detector Object Parameter File .................. 232
C.6 The Optics Object Parameter File... ........0......0.. 234
C.7 The Output Object Parameter File ................0.0.. 252
ix
List
Ll
2.1
2.2
2.3
2.4
2.9
2.6
2.7
2.8
2.9
2.10
2.11
2.12
2.13
2.14
2.15
2.16
2.17
3.1
3.2
3.3
3.4
3.9
3.6
3.7
of Figures
Physical representation of IRIMAGE simulator. ............ 7
Methods of backwards ray tracing... ..............04. 30
Radiating surface and receiving surface ................. 32
Relating detector area to emitting surface area. . 2... ....0.2.. 34
Representing and using the detector responsivity curve ........ 39
Noise model using probability density function. ..........2.. 43
Orientation of WCS with respect to the imaging system. ....... 46
Relating the GCS tothe WCS ... 0... es 48
Compositing Porche using VIO ..........0..0.0..02..004. 51
Example of the matte process ..........0..0...2..0000.2 53
Interpolating images on the Scene VGO...........0...... 55
2D representation of VGO construction using planes... . 2... 63
2D representation of a ray intercepting a 3D VGO........2... 64
Example of an 8x8 FPA using a 2x2 Unit Cell... 2... 2. 2. 67
The IRIMAGE optical system ..........0....0..02..0204 71
Detection of carbon monoxide(CO) ................042. 83
Images of Rotating Ellipsoid of CO ...........0.....0.. 84
Images of car with an exhaust cone of CO .......0.00.0022.. 86
Schematic of Amber AE-256 infrared imaging system ......... 94
Diagram of Amber AE-256 infrared camera... .........0.02.. 95
Varying range of experimental image to enchance picture ....... 100
Example of image calibration procedure... ...........0.02.. 102
Example of comparing range enhanced images ............. 103
Vibrational and rotational modes of CO .........0...0.0.00.. 107
3.8
3.9
3.10
3.11
3.12
3.13
3.14
3.15
3.16
3.17
3.18
3.19
3.20
3.21
3.22
3.23
3.24
3.25
3.26
3.27
3.28
3.29
3.30
3.31
3.32
A.l
A.2
A383
AA
Xl
Transmission curves of CO, COpg and CH, ...........02.202.. 112
RTE experimental setup .. 2... 0... 2 115
Temperature and emissivity maps for calibration plates ........ 119
Temperature/emissivity /matte maps for small heated disk ..... . 119
RTE images for AT = 10K... .....0.0..0.0..0..2. 2.20004 122
RTE images for AT = 5K .. 2... 2... ee 123
RTE images for AT =1K .........0.....0.....2.200. 125
Enhanced RTE images for AT = 1K ...............0..4 126
Methane experimental setup... 2... 0.0.0. ee ee ee 129
Example of CH, exhaust cone model .................. 134
Images of hot CH, plume against a hot background .......... 136
Images of hot CH, plume against a cold background. ......... 137
Images of cold CH, plume against a cold background ......... 139
Images of cold CH, plume against a hot background. ......... 140
Relationship between CO and COz: in engine exhaust ......... 142
The Gas Cell experimental setup. .. 2... 0... ee 143
Drawing of gas cell design ..........0.0...-20..00.0.0 008. 144
Schematic of gas mixing setup... 2... ee ee 145
Problem and solution to effects of gas cell windows .......... 152
Images of a cold CO2 plume against a hot background ........ 154
Images of a cold CO plume against a hot background ......... 155
Images of a cold CO2/CO plume against a hot background ...... 156
Images of a hot CO. plume against a cold background ........ 158
Images of a hot CO plume against a cold background ......... 159
Images of a hot CO2/CO plume against a cold background... .... 160
Block diagram of IRIMAGE ....................0..4. 168
Block diagram of Scene object... 2... ..00000..00..0004. 169
Block diagram of Atmosphere object .................. 170
Block diagram of Detector object .. 2... .0...0........... 171
xX
A.5 Block diagram of FPA object ........0...0..0...00 000. 172
B.1 Various VGO transformations ...................005., 176
B.2 Example of how data array is stored in the VGO ........... 179
B.3 Two methods of defining a bounded surface ..........2.2.2.. 184
B.4_ Example of normal connection method: Ellipsoid ........... 187
B.5 Example of alternating connection method: Box ............ 188
B.6 Example of a UPO interpolated ona 3D VGO ........20202.. 195
B.7 Cutting planes going along Z axis of the VGO ..........2... 196
B.8 Examples of multiple curve interpolation ................ 199
B.9 Two examples of a Torus interpolation ................. 200
B.10 Determining number of curves intercepted by a line segment .... . 201
B.11 Two curves interpolated ona 2D grid... ............... 202
B.12 Example of how IRIMAGE generates an atmospheric profile .... . 208
B.13 Various bounded volumes of CO in atmosphere ............ 210
xii
List of Tables
2.1 Amber IR camera system parameters ................0.., 81
3.1 The fixed parameters for the AE-256 camera... ........2.2.. 96
3.2 The fixed parameters used by IRIMAGE.............0022.. 98
3.3 Parameters for RTE experiments using filter NO3322-8 ........ 118
3.4 Parameters for RTE experiments using filter NO3990-4 ........ 120
3.5 Parameters for RTE experiments using an open filter (CaF: only) .. 120
3.6 Table of natural gas constituents ..................0. 129
3.7 Parameters for CH, experiments using filter NO3322-8......... 132
3.8 Parameters for CH, experiments using filter NO3990-4......... 132
3.9 Parameters for CH, experiments using filter NO4235-4......... 133
3.10 Parameters for CO & CO, experiments using filter N04235-4 ..... 149
3.11 Parameters for CO & CO: experiments using filter NO3689-4H .... 150
3.12 Parameters for CO & CO» experiments using filter N04693-4 ... .. 150
B.1 Parameters for building Ellipsoid and Box ............... 193
XIV
Glossary of Acronyms
AFGL
AVS
BLIP
CAD
CAE
CG
ECS
FITS
FLIR
FDA
FPA
FTIR
FWHM
GCS
GM
GMD
GUI
IR
NSF
OIA
OOP
POCS
Air Force Geophysics Lab
Advanced Visualization System
Background Limited Performance
Computer Aided Design
Computer Aided Engineering
Computer Graphics
Element Coordinate System
Flexible Image Transport System
Forward Looking Infrared imager
Flexable Data Array
Focal Plane Array
Fourier Transform Infrared spectroscopy
Full Width Half Maximum
Grid Coordinate System
Geometric Model
Geometric Model Database
Graphical User Interface
Infrared
Negative Space Filling
Object Index Array
Object Orient Programming
Primitive Object Coordinate System
QE
RTE
SOE
UPO
VGO
VIO
WCS
XV
Quantum Efficiency
Relative Temperature Experiment
Space Occupancy Enumeration
Universal Primitive Object
Versatile Grid Object
Virtual Image Object
World Coordinate System
Xvi
Glossary of ‘Terms
Atmosphere Object: A simulation object responsible for atmospheric modeling in
IRIMAGE.
Atmospheric Model: A set of parameters that define a specific set of atmospheric
conditions.
Detector Object: A simulation object responsible for modeling the four detector
types and the background fluctuation noise in IRIMAGE.
Element Coordinate System (ECS): A general coordinate system for a program-
ming object (such as a VGO or UPO) which makes internal operations easier
but can transform any location in itself to and from the WCS.
Flexible Data Array (FDA): Either a floating point or integer array that stores
the grid point data in the VGO.
FPA Object: A simulation object responsible for modeling each focal plane array
and driving the IRIMAGE simulation.
Geometric Model (GM): A three-dimensional representation of the surface of a
geometric object such as a cone or sphere.
Grid Coordinate System (GCS): The coordinate system for a VGO.
Object Index Array (OIA): An integer array that associates every grid point in
a VGO with the source of the data that it stores.
Optics Object: A simulation object responsible for modeling the optical system for
each FPA in an IRIMAGE simulation.
Output Object: A simulation object responsible for outputing the image data for
each FPA in an IRIMAGE simulation.
xvii
Program Object (PO): A programming class or structure that defines a complete
and self-contained piece of programming code.
Scene Object: A simulation object responsible for modeling the background sources
in IRIMAGE.
Simulation Object (SO): An abstract element of the simulation such as the Scene
object or FPA object.
Universal Primitive Object (UPO): A programming object capable of modeling
a bounded surface using a hybrid polygon-edge based surface model. Specifically
designed for interpolating onto a VGO.
Versatile Grid Object (VGO): A program object that defines a multi-dimensional
orthogonal data array which stores spatially dependent data (e.g., image and
volumetric data).
Virtual Image Object (VIO): A program object that stores a set of image trans-
formations which define the location, orientation and scale of the image on the
Scene object’s imaging VGO.
World Coordinate System (WCS): The coordinate system for the entire IRIM-
AGE simulation. Located at the front surface of the optical system. All other
coordinate systems must be able to transform points to and from the WCS.
Chapter 1 Overview of Research
This thesis discusses the research related to the development of an infrared imaging
system simulator and its application to investigations into passive, multi-spectral,
infrared imaging. The potential of passive, multi-spectral imaging for environmental
monitoring and other scientific and commercial applications has long been a matter
for discussion and debate. The development of experimental systems to explore these
multi-spectral applications has been hampered by the high cost of such systems and
the potentially small market they may represent. Therefore, it was the intention of
this project to address the following issues related to multi-spectral infrared imaging:
e What are the possible commercial applications of infrared imaging, specifically
multi-spectral infrared imaging?
e Is it possible to use passive, multi-spectral, infrared imaging to detect the pres-
ence of gaseous pollutants?
To answer such specifically systems oriented questions it is necessary to inquire into
two more general issues.
e Is it possible to develop a cost-effective method of defining a standard scene
and use it to compare the performance of different infrared imaging systems?
e Can we apply the same method to judging the performance of a particular
imaging system that looks at a variety of different scenes?
We investigated these issues by developing a flexible computer simulation of the in-
frared imaging process and by conducting a series of multi-spectral experiments. The
results of this investigation are an attempt to provide answers and solutions to the
questions raised.
As a result of this investigation, this thesis describes several contributions we have
made to the infrared imaging community. One of these contributions is the develop-
ment of a flexible computer simulation, called IRIMAGE, that simulates the entire
imaging process from the background scene to the output from an imaging system.
Because it simulates the entire imaging process, IRIMAGE is capable of comparing
different imaging systems against a standard scene and comparing a specific imag-
ing system against a variety of scenes. IRIMAGE can also serve as an application
research tool, investigating different applications of single band and multi-spectral
infrared imaging. Using IRIMAGE in this capacity leads to the other significant
contribution of our research. In an effort to validate IRIMAGE and investigate
passive, multi-spectral detection of gaseous pollutants, we conducted a set of ex-
periments that imaged various plumes of CH4, CO, and CO», against either a hot
or cold background. These images were generated using an infrared camera fitted
with a filter wheel containing filters to isolate the emission bands of these pollutants.
Subsequently, we simulated the approximate conditions of these experiments using
IRIMAGE. By comparing the images from the experiments with the simulated im-
ages, we are able to demonstrate the validity of IRIMAGE. More importantly, we
prove that it is possible to isolate and detect these three gases independently using
passive, multi-spectral infrared imaging. However, we also determined that certain
technological and performance issues must be resolved before this is a viable method
of pollution detection. The rest of this chapter details the motivation for this project,
discusses the IRIMAGE simulation, and presents an overview of the multi-spectral
experiments discussed above.
1.1 Motivation
With the discovery of the infrared region of the electro-magnetic spectrum by Sir
William Herschel in 1800 and the subsequent discovery of the absorption and trans-
mission band nature of infrared radiation by his son, Sir John Herschel, in 1840, a new
era in non-visible, optical investigations began. Since its humble beginnings as a re-
search tool, the generation, detection and imaging of infrared radiation has become a
multi-million dollar industry that. encompasses a vast array of military (night scopes,
targeting systems, etc.) and commercial (communications, astronomical observations,
motion detectors, etc.) applications. Although a broad array of commercial applica-
tions for single detectors abound, very few imaging applications exist. Furthermore,
it is becoming increasingly difficult for the industry to bear the cost of supporting
efforts to commercialize and improve infrared imaging technology during this current
era of government downsizing. Therefore, the infrared imaging community needs to
find ways to increase the commercial market for infrared imaging while, at the same
time, reducing the costs of manufacturing and supporting current technology as well
as developing new technologies.
We believe that many of the applications of infrared imaging will require or be
enhanced by some form of multi-spectral capability. The advantages of multi-spectral
infrared imaging are analogous to the advantages of color images over black and white
images. While a black and white image provides only the intensity of light observed, a
color image provides both chromatic and intensity information. The additional chro-
matic information adds to the total information known about an object being imaged.
Similarly, a multi-spectral infrared image contains additional information which also
improves our ability to discern the unique characteristics of an object that would
otherwise not be obtained from a single-band infrared image. These characteristics
include the temperature and emissivity of a surface or the transmission and emission
bands of a gas. Active imaging measures the change in the incident radiation of a
known source due to the absorption of a substance. For instance, the most common
use of active multi-spectral infrared imaging is terrestrial monitoring [1, 2]. These
images are captured by airbourne or space-bourne, high resolution, imaging spec-
trometers (32 to 224 bands) which collect the reflected sunlight (.4 um- 2.45 pm)
from the earth’s surface. From these images, we are able to differentiate minerals
in the earth’s crust; determine ocean temperatures; discriminate between different
types of vegetation; and monitor certain characteristics of the atmosphere. Passive
multi-spectral detection assumes that the objects of our investigations are either the
source of the infrared radiation or attenuate a background source whose own radiative
characteristics are not known. Although most of the standard infrared imaging sys-
tems are passive imagers, none of these systems are capable of real-time multi-spectral
imaging. However, we believe there are certain applications such as temperature mea-
surement, environmental monitoring and medical diagnosis which could benefit from
the development of passive, multi-spectral infrared imaging systems [3].
As mentioned above, one of the possible applications of passive multi-spectral IR
imaging is environmental monitoring, specifically pollution detection. As concerns
about global warming and the enforcement of government mandated air quality stan-
dards increase, there is an overwhelming need for remote sensing and imaging of
both natural and man-made gaseous emissions. Current methods of remote sensing
of gaseous pollutants involve active detection such as FTIR (Fourier Transform IR)
spectroscopy [4], laser heterodyne radiometry [5] or longpath photometry [6]. The
drawback to these active methods is that they require some type of IR source which
may not be convenient or even possible in certain situations. In addition, none of
these methods image the emissions, but rely on the pollutants to pass through the
active region between the source and the detector. On the other hand, since most
gaseous pollutants are emitted at an elevated temperature, it may be possible to pas-
sively detect the emission spectra of these gases. A passive, chemical vapor detection
system involving a FLIR (Forward Looking IR imager) with several narrow-bandpass
filters and a series of detection algorithms was proposed in 1991 by Althouse and
Chang [7]. Although their application was military based, they surmised that such
a system could also be used in environmental pollution detection. Our experiments
and simulations of passive, multi-spectral pollution detection rely on similar filtering
concepts.
One increasingly popular method used to design and test systems is simulating
how that system will perform under various conditions. In this approach, the entire
infrared imaging system is modeled on a computer. Then using a set of test scenes
(background sources and atmospheric models), the viability of this system could be
compared with other existing imaging systems or new system designs. In addition,
different aspects of a system can be adjusted to find the best balance of functionality
and performance. Such an approach has been very effective in the design of electrical
devices, integrated circuits and optical systems using simulators like Spice3 [8], ANA-
LOG and BEAM. By testing new designs prior to their fabrication, it is possible to
minimize the time and cost of the “trial and error” phase of development. In addition,
a simulator gives the system designer the freedom to investigate designs that might
be too expensive or difficult to fabricate without evidence to support the design.
Besides aiding in the R&D of an infrared imaging system, the same simulation
could be used to investigate new applications of current and future infrared imag-
ing technology. Using the information generated by such a simulation, a systems
designer could fine tune existing systems for current applications or determine the
best technology for a new application. It would also be possible for a researcher to
investigate the performance requirements of a proposed application. Ultimately, a
computer simulation of the entire imaging process could be used to tailor a complete
imaging system for any application. Therefore, a flexible simulation capable of both
imaging system design and infrared application research would not only reduce costs
but would help stimulate the commercial market with new applications of infrared
imaging technology.
Although there are several simulations related to infrared imaging and detection,
none of these simulators are comprehensive enough to satisfy the need for a complete
infrared imaging simulation. Several scene simulations exist including GTVISIT by
Cathcart and Sheffer [9] and DIRSIG by Schott, Raquefio and Salvaggio [10]. These
models use first principle computations and three-dimensional models to simulate
the infrared radiation emitted by a scene. Despite their sophisticated modeling of
background scenes, these simulations fail to properly model a dynamic and hetero-
geneous atmosphere and use a very idealized imaging system. On the other hand,
several detector simulations like Stanford’s SUMerCad [11] and Dawn Technologies’
SEMICAD DEVICE [12] model the response of individual detectors. These simula-
tions do not attempt to model a background scene, but use a standard gray body
source for analysis instead. Between these two extremes lie several simulations that
do attempt to model both the imaging system and the background scene [13, 14, 15}.
However, these simulations are designed for either a specific application or a specific
imager type. Although these simulations model the atmosphere using the Air Force
Geophysics Lab (AFGL) LOWTRAN/MODTRAN atmospheric code [16], they only
take advantage of its layered nature making the atmospheric model static and purely
one-dimensional. In one way or another, each of these simulations falls short of the
goal of a design and applications research tool capable of modeling the entire infrared
imaging process. They either do not incorporate all of the major parts (source, at-
mosphere, optics and imaging method) or are designed for a specific type of imaging
system or application.
1.2 The IRIMAGE Simulation
IRIMAGE bridges the gap between the scene generation programs and the detector
simulations. It provides both a tool for designing and comparing different imaging
systems and an application research tool for investigating the various applications of
infrared imaging. Using a simple ray tracing technique in conjunction with a well
defined background scene, atmospheric model and a simple optical model, IRIMAGE
computes the infrared radiation that is incident upon the surface of each detector
in a focal plane array. Each detector uses its responsivity curve and a background
fluctuations model to convert the incident radiation into a signal, noise and total
voltage. Following the object-oriented programming paradigm, IRIMAGE organizes
these operations into a series of self contained objects that represent the major ele-
ments of the infrared imaging process. Data is passed between these objects in such
a manner that improvements to any one of the program objects does not effect the
other elements of the simulation. The parameters that define each element of a sim-
ulation are stored in a set of parameter files which are generated by the Graphical
User Interface (GUI). All of these pieces put together result in a flexible simulation
package that can generate results now and can be improved in the future.
Figure 1.1 is a physical representation of IRIMAGE. As this figure points out,
Scene Object
Image _— Imaging . . FPA Object
Formation _ Plane ( Optics Object &
Detector Object
Atmosphere Obie 4%
ODTRAN;
m ) ( Output Object)
Figure 1.1: Physical representation of IRIMAGE simulator.
there are several major elements that make up the simulation. IRIMAGE attempts
to treat each of these major elements as separate objects. By doing so, we are able
to easily control and modify each of these elements independently.
The first element, or simulation object, of IRIMAGE defines the temperature and
emissivity of background infrared sources. The Scene simulation object represents
the non-atmospheric sources of the scene using a two-dimension background surface.
This background surface is a two-dimensional grid called the imaging VGO (Versatile
Grid Object). The imaging VGO allows us to reduce the source objects to a series of
two-dimensional temperature and emissivity maps that may be interpolated onto the
imaging VGO. Using a manipulation object called the Virtual Image Object (VIO),
each of these maps can be translated, rotated and scaled in the 2D plane before being
interpolated. Furthermore, these transformations can be animated using a series of
key frames to store the manipulations for specific frames. Using the key frames, it
is then possible to interpolate the transformation values between two key frames. In
addition, multiple images can be overlaid on the imaging VGO using a simple “matte”
to define what part of each image to interpolate. This two-dimensional approach is a
first order approximation of a full three-dimensional scene simulation. Since such 3D
simulations already exist, our intention was to provide a simple method of modeling
a scene while being flexible enough to be able to incorporate an existing 3D scene
generator in the future.
The other element responsible for modeling the scene is the Atmosphere simulation
object. ‘This element models the atmosphere between the Scene object and the imaging
system. In this case, the atmosphere is modeled using a three-dimensional form of
the VGO. The atmospheric VGO is used to define specific regions of the atmosphere
that may have different characteristics such as a polluted plume or a hot gas cloud.
Each of these polluted regions is represented by a bounding surface model, a gas
model and an aerosol model. The parameters for each gas and aerosol model are
stored in a set of databases. The bounding surface is linked to these models via their
database index. Using an interpolation algorithm we developed, the Atmosphere
object. interpolates each bounding surface onto the VGO. By doing so, every grid
point inside the bounding surface is then assigned the surface’s corresponding gas
and aerosol model index. Like the images in the Scene object, each of these bounding
surfaces may also be animated using the key frame procedure defined above. Once
all of these surfaces have been interpolated for a given frame, the Atmosphere object
uses another interpolation alogrithm to generate the atmospheric profile along each
ray that passes through the VGO. It combines this atmospheric profile with the
_ temperature and emissivity values of the ray’s interception point on the imaging VGO
and passes this data to the MODTRAN object, a modified version of the atmospheric
model developed by the Air Force Geophysics Lab. The MODTRAN object generates
the spectral radiance array for the ray using the ray’s background source information
and atmospheric profile. Once the spectral radiance array is computed, it is passed
onto the imaging system.
The IRIMAGE imaging system is made of four major elements: the Focal Plane
Array(FPA) object, the Optics object, the Detector object, and the Output object. The
FPA simulation object acts as the overall control object for the imaging system. It
lays out the detectors onto the focal plane by tiling a unit cell across the array. It also
generates the rays used to compute the radiance incident on each detector and passes
them onto the other major elements. Once the radiance is computed, the FPA object
passes this information onto the Optics object and then the Detector object. Finally,
the FPA retrieves the output voltages, photon power and photon arrival rates from
the Detector object and passes them onto the Output object to be output into a file.
IRIMAGE is also capable of defining multiple FPA’s that look at the same scene. In
some cases, this means that each FPA may need its own Optics object and Output
object. Therefore, these objects are defined to be part of the FPA object.
The Optics simulation object defines the optical system of the imaging system.
Currently, we use the pin-hole approximation in conjunction with the nodal points of
the optical system. This approximation uses the concept that a ray passing through
one of the nodal points of the system will appear to exit from the other nodal point
with the same direction vector. Therefore, the Optics object stores the location of
the cardinal points of the optical system (focal, principle and nodal points) and the
location of the front and back surfaces along the optical axis. In addition, it stores
the F-number (F’/#) of the system and maintains a database of transmission curves
for different filters used in the optical system. When a ray is generated by the FPA, it
is passed to the Optics object. It determines the point where the ray exits the optical
system, adjusts the ray and returns it to the FPA object. Once the spectral radiance
array is calculated for the ray, this array is passed back to the optical system which
then uses the filter transmission curves to attenuate the incident radiance. After this
attenuation is complete, the Optics object waits for the next ray.
The individual detector models are stored in the Detector simulation object. This
object maintains a database of detectors which are based on a general detector model.
Currently, we model four different detector types: Photoconductors, Photovoltaics,
Pyrometers, and Bolometers. Although the physical processes involved with detection
may be different, each of these detectors can be represented by a spectral responsivity
curve. This curve defines what the resultant voltage is for an incident number of
10
photons of a particular frequency. Since the radiance data is also given in the form of
photons per frequency interval, this approach is an excellent method for generalizing
each detector system as well as computing the output voltage that results from a
given radiance array. This object also defines a model for generating the random
noise associated with the background photon fluctuation. This time-dependent noise
model approximates the random noise due to the random fluctuations in the incident
number of photons for each frame. This noise model combined with the responsivity
curve representation generates the signal, noise and output voltages for a particular
detector.
Finally, the last element of IRIMAGE is the Output object. This object simply
takes the output values from the Detector object and puts them in an ASCII data
file. Each frame of the simulation is stored as a separate file. The object-oriented
nature of this object will allow future versions of IRIMAGE to output the data in
various image formats as well as onto the screen.
The parameters that define each of the major elements of the IRIMAGE simulation
are stored in a set of parameter files. The Graphical User Interface (GUI) generates
a parameter file for each of the six major objects of IRIMAGE. In addition to the
six parameter files for these objects, the GUI also generates a general parameter file
which sets up a few of the animation parameters, the monitoring flags, and links
each major object with its own parameter file name and location. Each of these
parameter files are written in ASCII format so they are portable to virtually any
operating system that IRIMAGE supports. We avoid a direct link between the GUI
and IRIMAGE because we prefered to be able to run our simulations in batch mode
(multiple simulations at a time) which is difficult to implement if the GUI is linked
directly to the simulation.
The actual simulation process in IRIMAGE is very straightforward. It begins by
loading and initializing each of the major objects. Once each object is prepared, the
imaging and atmosphere VGO’s are set up for the current frame. With the scene and
atmosphere prepared, the FPA generates each ray and passes it to the Optics object to
compute the exit ray. This ray is then passed to the Scene and Atmosphere objects to
ll
generate the profile and then compute the spectral radiance array using MODTRAN.
This spectral radiance array is passed back to the Optics object to be attenuated and
is then passed on to the Detector object. Using the detector associated with the given
detection element, the Detector object computes the signal, noise and total voltages.
These values are passed onto the Output object which loads them into the ASCII file
for the current frame. This process is repeated for each detector in the FPA and each
frame of the simulation.
1.3. Multi-Spectral Infrared Imaging Experiments
The experimental phase of this project had two primary goals. The first goal was to
investigate the viability of passive pollution detection using multi-spectral methods,
one of the primary issues that we intended to investigate further with this project.
The second goal was to verify the results of the IRIMAGE simulation using a set of
simple experiments that test the various parts of IRIMAGE. In an effort to codify the
verification process, we laid out three criteria that IRIMAGE should satisfy in order
to be considered a valid simulator:
1. The Scene object accurately emulates a simple background scene that has re-
gions with varying temperatures.
2. The Atmosphere object reasonably approximates the absorptive and radiative
effects of a three-dimensional heterogeneous atmosphere.
3. The Imaging system (made up of the FPA, Optics, Detector and Output ob-
jects) generates a reasonable and realistic output based on the actual physical
parameters of the experimental setup.
Using these criteria and the desire to investigate passive pollution detection, we de-
signed three experiments that would satisfy both goals. Each of these experiments
uses an Amber AE-256 infrared camera consisting of a 256x256 InSb array and a five
position filter wheel in a LN» dewar. By using several different narrow band-pass fil-
12
ters in the filter wheel, we were able to incorporate multi-spectral imaging into these
experiments.
The first of these experiments was the Relative Temperature Experiment (RTE).
Designed to test the first and third validation criterion, the RTE consisted of a large
aluminum plate and a smaller aluminum disk located in the center of this plate. The
small disk was thermally isolated from the large plate using styrofoam and had a
small heater attached to its backside. The surfaces of both plates were painted with
a high emissivity black paint to maximize the generated infrared radiation. Using
the heater, we controlled the temperature difference between the large plate (at room
temperature) and the heated disk. We used a thermocouple attached to both plates
to measure their temperature difference and used a standard surface thermocouple to
measure the absolute temperature of the large plate. Using several different filters,
we captured images of various temperature differences. The inherent two-dimensional
nature made this source ideal for testing the 2D source modeling in the Scene object.
For the simulation, the measured temperature data is applied to a set of images that
approximate the large plate and the small disk. Using these images and each filter’s
transmission curve, IRIMAGE generated a set of simulated images for each filter.
The results of these simulations were images that were very close to the experiment’s
images, with some minor variance from the experiment. This variance may be at-
tributed to temperature variance across the surface of the large plate (especially near
the disk/plate interface) which could effect the relative and absolute temperature
measurement. Since this behavior was not consistent between filters, we also suspect
that the approximations made when digitizing the filter transmission curves may also
be affecting the image, especially in the narrow bands. Overall, we felt IRIMAGE
easily satisfied both the first and third criteria for validation.
Although the first experiment was exclusively for verfiying IRIMAGE, the second
experiment had broader implications. Referred to as the “methane experiment,”
this experiment introduced a plume of highly concentrated methane (92.88%) and
a trace amount of CO, (.22%) into a clean atmosphere against a fixed temperature
background. The plume was generated by passing natural gas from the laboratory gas
13
line through a copper coil placed in a water bath. By adjusting the water temperature,
we controlled the temperature of the plume. Using the absorption/emission spectra
of CH, and COnz, we selected three filters to isolate the band-pass of the camera to the
methane band, the carbon dioxide band and a clean band (not containing either gas).
For each filter we examined the four possible cases of hot and cold backgrounds versus
hot and cold plumes. We obtained the gas concentration data from the gas company
and recorded the temperatures of the plume and clean atmosphere as well as the
background surface for each case and filter. From this data and the camera setup data,
we conducted a series of simulations of these experiments as well. The most important
result was the ability to passively detect methane and COg in their respective spectral
bands. In addition, we did not detect any gas in the clean region, which implies that
the plume was not part of the background plate and that the CH, and CO» bands
were not overlapping. The simulation results were as encouraging. The simulated
images compared very well to the experiment images. The major variances occurred
when the cone that approximated the plume grew wider but still maintained the
same concentration (not physically correct). This caused this simulated region to
appear brighter when it was imaged against a cold background and vice versa against
a hot background. However, near the tip of the exhaust nozzle, the images looked
very similar. In addition, the background sources and noise modeling appeared to be
much more consistant than in the RTE images. The excellent corroboration between
experiment and simulation proves that IRIMAGE satifies the second and third criteria
for validity. Furthermore, the experimental evidence and simulated images not only
demonstrate the viability of detecting a highly concentrated gas, but also the ability
to image a very minor consituent of that gas as well.
The third experiment proved to be the most difficult. In this experiment, we
chose to examine a real world application of pollution detection using multi-spectral
methods. Using a setup similar to the methane experiment, this experiment gener-
ated plumes nitrogen containing 10% CO», 1% CO and a mixture of 6.7% CO2/ 1%
CO. The poisonous nature of carbon monoxide (CO) necessitated the use of a gas
cell that would isolate the polluted atmosphere from the laboratory atmosphere. The
14
constraints of the imaging system, in combination with the desire to minimize depth
of field problems, forced us to build a large gas cell which required large windows
that were transmissive in the infrared. Unfortunately, the only windows transmissive
enough to allow the detection of the plumes were made of poly-ethylene based plastic
wrap. Although very transmissive, the non-uniformity of the surface caused noticable
differences in transmission across the surface of the window. Therefore, we incorpo-
rated an image processing technique to try to negate the effects of these windows. The
resulting images showed marginal improvement, but introduced artifacts that seemed
to wash out most of the expected detail. Again, we used three filters to isolate the
band-pass of the camera to the CO2 band, the CO band, and a clean band. However,
in this experiment we decided to examine only the cases of a hot plume against a cold
background and cold plume against a hot background. For each filter and case, we
recorded the temperatures of the plume, the clean atmosphere and the background
surface. We also recorded the concentration of the gases in the plume. Using this
data and the camera data, we defined the simulations for these experiments.
Unlike the methane experiment, the experimental results of the gas cell experi-
ments were mixed. For the hot background/cold plume case, we were able to detect
the carbon dioxide in a still frame, but carbon monoxide was much fainter and could
only be seen by viewing successive frames. For the other case, we were able to see
the plumes for each gas, but again very faintly. We attribute these problems to the
problems with the windows. In all cases, the clean band showed no trace of either
gas. Comparing the simulations to the experimental images, the simulation predicted
that the plumes would be more visible. The results of the methane experiment agree
with these simulations as well since the concentration of COz in that experiment was
between 25 and 50 times smaller and that plume was still visible. In general, these
experiments demonstrate the possibility of independently detecting CO and CO, us-
ing passive, multi-spectral IR imaging. Since the results of the simulations predict
that both plumes should be easily detected, we believe further experiments should be
conducted using a better experimental apperatus.
15
Besides the detailed conclusions about each experiment and its relationship to
IRIMAGE, there are some general conclusions that can be made. First, using a system
like the AE-256 has significant problems as a true multi-spectral imager. Since the
camera has to be recalibrated for each filter, it is impossible to generate near-real time
multi-spectral images using this system. It might be possible if each of the filters in
the wheel had central wavelengths and bandwidths that correspond to approximately
the same number of collected photons over the desired temperature range, but this
would put severe limitations on these filters. Another solution might be to use a set
of microfilters with each filter sitting over a single detector. Such an arrangement
would allow the camera to be calibrated for all of the bands simulataneously as
well as provide true real time multi-spectral imaging. However, if the scene has
a lot of spatial variance, there will be problems where two adjacent pixels may be
looking at two very different sources. Such a solution could be ineffective for pollution
imaging. Therefore, the further technological advances and research are neccessary
before passive, multi-spectral infrared imaging is economically feasible.
Another conclusion from these experiments is that although IRIMAGE generates
reasonably accurate images, there is still room for improvement. The atmospheric
modeling in the methane experiment demonstrates the fundemental problem of mod-
eling an expanding gas in IRIMAGE. As the bounding surface grows, the concen-
tration remains the same. Although this is physically impossible, it is possible to
simulate a series of sucessive surfaces that do decrease in concentration as they get
larger (our two cylinder gas for example). However, it would be better if IRIMAGE
were to define a method of diluting the gas inside a volume as it grows. Such a
solution would require certain modifications to be made to MODTRAN as well as
IRIMAGE. Another place for improvement is detector noise modeling. Initially, we
assumed that a detector would be operating under background-limited conditions
(BLIP). This is a reasonable approximation when looking at a wide spectral band,
but as the band is narrowed for multi-spectral imaging, the detector noise begins
to dominate the system. This would especially be true at shorter wavelengths (less
than 4jm) were the radiation of a near room temeprature source gets considerably
16
smaller than for the longer wavelengths. In these experiments, the effects of the sim-
ulated noise model on the simulated images appeared to be much smaller than in the
corresponding experimental images for the filters centered around 3.5um and having
very small bandwidths (~ .07um). In these cases, the experimental images looked
considerably noisier than the simulated counterparts. Although these improvements
would make IRIMAGE an even better simulation, the results of these experiments
prove that IRIMAGE is already a valuable design and application research tool.
1.4 Outline of Thesis
The rest of this thesis discusses the concepts and experiments introduced in this
chapter in greater detail. Chapter 2 provides a detailed discussion of the IRIMAGE
simulation. This includes a comprehensive discussion of the physical basis and com-
putational methods used by IRIMAGE. It also expands on the six major elements
of the simualtion and how they fit together. Finally, it presents the results of two
example simulations.
Chapter 3 presents a comprehensive discussion of the experiments used to inves-
tigate passive multi-spectral pollution detection and validate the simulation. The
chapter includes a complete description of the Amber imaging system and how the
IRIMAGE simulator modeled this system. There is a brief overview of multi-spectral
pollution detection. The three following sections describe each experiment and make
comparisons between the experimental and simulation results.
There is also a three part appendix. Part A provides a set of block diagrams
that show how the different objects in IRIMAGE are related. Part B is a very
detailed discussion of the computational methods employed to interpolate bounded
volumes onto a versatile grid object. This discussion expands on concepts introduced
in chapter 2. Part C is an example of the parameter files used by IRIMAGE to
simulate the car example in chapter 2.
Finally, a web site exists to support the IRIMAGE software. The latest versions
of the source code and the GUI are available at www.ssdp.caltech.edu.
17
Bibliography
[1]
A.F.H. Goetz, “Imaging spectrometry for Earth Observations,” Episodes, 15(1),
March 1992, 7-14.
For an excellent discussion of terrestrial monitoring using imaging spectroscopy:
G. Vane and A.F.H. Goetz, “Terrestrial Imaging Spectroscopy,” Remote Sensing
of Environment, 24, 1988, 1-29.
S. Hejazi, D.C. Wobschall, R.A. Spangler, and M. Anbar, “Scope and limitations
of thermal imaging using multiwavelength infrared detection,” Optical Engineer-
ing, 31(11), Nov. 1992, 2383-2393.
H. Phan and J. Auth, “Measurements of chemical emissions using FTIR spec-
troscopy,” American Laboratory News, Aug. 1993
D. Courtois, D. Delahaigue, and C. Thiebeaux, “Detection of thermal emission
from atmospheric gases by laser heterodyne radiometry,” Infrared Physics, 34(1),
1993, 407-413.
G.A. Bishop, J.R. Starkey, A. Ihlenfeldt, W.J. Williams, and D.H. Stedman,
“IR longpath photometry, a remote sensing tool for automobile enissions”, Anal.
Chem., 61(10), Oct. 1989, pp. 671A.
M.L.G. Althouse and C.I. Chang, “Chemical vapor detection with a multispectral
thermal imager,” Optical Engineering, 30(11), Nov. 1991, pp. 1725-1733.
Spice was developed by the Computer-Aided Design Group, Department of Elec-
trical Engineering and Computer Sciences, University of California-Berkeley. For
more information, contact them via the web at www-cad.eecs.berkeley.edu.
A.D. Sheffer and J.M. Cathcart, “Computer generated IR imagery: a first prin-
ciples approach,” Proc. SPIE, 933, 1988, 199-206.
[10]
[11]
[12]
[13]
18
J. R. Schott, R. Raqueno, and C. Salvaggio, “Incorporation of a time-dependent
thermodynamic model and radiation propagation model into infrared three-
dimensional synthetic image generation,” Optical Engineering, 31(7), July 1992,
1505-1516.
J.L. Meléndez and C.R.Helms, “Process Modeling and Simulation for
Hg;..Cd,Te. Part I:Status of Stanford University Mercury Cadmium Telluride
Process Simulator”, Journal of Electronic Materials, 24(5), May 1995, 565-572.
For more information contact: Dawn Technologies Inc., 491 Macara Avenue,
Sunnyvale, CA 94086.
W.T. Kreiss, A. Tchoublneh, and W.A. Lanich, “Model for infrared sensor perfor-
mance evaluation: applications and results,” Optical Engineering, 30(11), Nov.
1991, 1797-1803.
H.V. Kennedy, “Modeling second-generation thermal imaging systems,” Optical
Engineering, 30(11), Nov. 1991, 1771-1778.
G.H. Kornfeld, “Digital simulation of precise sensor degredations including non-
linearities and shift variance,” Proc. SPIE: IR Image Process. and Enhance., 781,
1987, 63-70.
F.X. Kneizys, L.W. Abreu, G.P. Anderson, J.H. Chetwynd, E.P. Shettle, A. Berk,
L.S. Bernstein, D.C. Robertson, P.K. Acharya, L.S. Rothman, J.E.A. Selby,
W.O. Gallery, and S.A. Clough, “The MODTRAN 2/3 Report and LOWTRAN
7 Model,” prepared for PL/GPOS, 1996.
19
Chapter 2 IRIMAGE: An Infrared
Imaging System Simulator
2.1 Introduction
Since the beginning of the computer age, scientists and engineers have used computer-
based simulations to investigate numerous theories, designs and ideas. Initially, simu-
lations performed the computational drudgery of numerical analysis and other simple
algebraic calculations. As time progressed and computers became more sophisticated,
more advanced numerical methods could be employed to solve linear differential equa-
tions. With advent of Computer Graphics (CG), the simulation community grew
exponentially. By adding the visual component to simulations, one could “view”
a simulation’s results and quickly make adjustments or even adjust the parameters
while the simulation was still running. With the advent of cheaper and more power-
ful computers, the concept of Computer Aided Engineering (CAE) took hold. Using
CAE, solutions to real world problems could be tested on the computer, saving the
time and money that might be spent attempting to find the solution in other ways.
Computer Aided-Design (CAD) evolved from the CAE model. Using a CAD system,
scientists and engineers could completely design and conduct simple tests on various
models or equipment. Not only did this save time and money, it allowed a greater
sense of freedom. A scientist could explore various cases without spending months
adjusting an experiment. An engineer could not only tailor a specific product to
an application but also look for alternative solutions to a particular problem. With
acceptance of the Object-Oriented approach to problem solving and programming,
new CAD systems allowed users to pick and choose what elements to incorporate
in their models and which ones to disregard. It made creating CAD systems easier
and improved the ability to tailor a CAD system to a particular application without
20
losing its flexibility and maintainability. Using this new approach, we have created
the IRIMAGE Infrared System Simulator to provide an object oriented CAD model
for designing, testing and experimenting with various IR imaging systems.
In the past, IR system simulation has been approached from two directions. The
first was inspired by the desire to simulate IR backgrounds and scenes. Programs like
those created by Schott, Raqueno and Salvaggio [1] or by Cathcart and Scheffer [2]
were created to use 3-D scene information and generate images that would represent
what an IR scene would look like under various conditions. Although these simula-
tions use sophisticated computer graphics models and first principle calculations to
model the final scene, the atmospheric modeling is not as sophisticated, usually using
standard atmospheric models like those in LOWTRAN [3]. In addition, the imag-
ing system is largely ignored for simplicity sake or is treated as an image processing
task. While these approaches certainly simulate what a scene would look like to the
perfect imager looking through a standard atmosphere, they do not adequately ap-
proach how one would actually generate such images using an imaging system. The
second approach involves the simulation of various detection devices. Programs like
SUMerCad [4] or SEMICAD DEVICE by Dawn Technologies [5] simulate device per-
formance using the physical parameters of a device in combination with equations
for various detector types and material systems. The results would be the expected
performance of the detector. Although a very effective method for testing detector
performance, these simulations do not test the detector in an imaging system or test
it against a realistic background.
IRIMAGE bridges the gap between the scene generation programs and detector
simulations. It generates a simple background scene and uses a backward ray tracing
approach to determine the incident IR radiation on the optical system of the imaging
system. Using the spectral character of the incident radiation, the simulation com-
putes the output signal and noise of an FPA based on the response curve of each
detector. Thus, IRIMAGE generates output images of the background scene and
intervening atmosphere using an actual imaging system.
The IRIMAGE simulator uses a straightforward Object-Oriented Programming
21
(OOP) approach instead of the classical linear programming methodology. Under
the OOP paradigm, a basic object contains both data structures to store informa-
tion and routines which can manipulate this information. Some basic objects define
standardized methods for storing and manipulating specific types of data, i.e., vec-
tors, matrices, grids, etc., while others are more specialized to perform specific tasks.
Larger objects are constructed from these smaller objects while incorporating new
variables and routines as well. Eventually, the program becomes a collection of cou-
pled, self-contained objects that pass information, or other objects, between each
other in order to perform the various tasks of the simulation. In terms of IRIMAGE,
the OOP approach allows us to treat the various physical elements of the infrared
imaging process as “objects” which interact with each other to generate an output
image.
In chapter 1, figure 1.1 presented a physical representation of IRIMAGE. As this
figure points out, there are several major elements that make up the simulation.
IRIMAGE tries to treat each of these major elements as a separate object. By doing
so, we are able to easily control each of these elements independently.
The first two elements of the simulation generate the IR source information. The
background Scene Object defines a 2D background surface for the scene. The surface
represents objects in the background by projecting various temperature and emissivity
maps onto an image plane represented by a 2D grid approach. The atmosphere
between the Scene object and the imaging system is represented by the Atmosphere
Object which models the 3D nature of the atmosphere using a three-dimensional grid
and the MODTRAN atmospheric model. IRIMAGE uses a backward ray tracing
method in conjunction with these grid-based objects to compute the spectral radiation
incident upon the optics of imaging system.
The IRIMAGE imaging system is made of four major elements. The Focal Plane
Array (FPA) Object acts as the overall control object for the imaging system. It
lays out the detectors onto the focal plane; generates and passes on the rays used
to compute the radiance for each detector; passes the incident radiance information
to the optics and detector objects; and passes the output voltages to the output
22
object. With this in mind, each FPA object is associated with an Optics Object
which stores the parameters which define the optical system. The Optics object
manipulates the rays generated by the FPA and attenuates the collected radiance
generated by MODTRAN. The FPA passes the radiant power returned by the Optics
object to specific detectors stored in the Detector Object. Using the radiant power
and the responsivity of the corresponding detector(s), the Detector object generates
the signal output as well as the noise output which uses a noise modeling algorithm.
Once the FPA receives the output from the corresponding detector element, it passes
the data to the Output Object to be written out to a data file.
The parameters that define an IRIMAGE simulation are loaded by each of these
objects using a set of seven parameter files generated by a separate Graphical User
Interface (GUI). When IRIMAGE begins, it loads the general parameter file which sets
up a few of the animation parameters and the monitoring flags. In addition, it links
each major object with is own parameter file which contains all of the information
necessary to define the corresponding object. These parameter files are written in
ASCII format so they are portable to virtually an operating system that IRIMAGE
supports. We briefly discuss the GUI and these scripts at the end of the paper.
Besides the obvious conceptual advantage of creating a program like one builds a
house, the other advantages to the OOP approach include code reuse and minimal
code upkeep. Since many of the basic objects used by IRIMAGE perform standard
tasks or define standard data structures and operations, their code can be reused over
and over by the various larger objects in the simulation. Several examples of code
reuse are discussed in the paper. More importantly, most of the objects in IRIMAGE
are self contained. Thus, changes in one object do not require changing the entire
program which minimizes the upkeep of the code and provides maximum flexibility
in the future for upgrades. In a simulation like IRIMAGE, this benefit is essential to
the long term usefulness of the program.
Before discussing the simulation itself, we briefly discuss the major physical princi-
ples utilized by IRIMAGE in section 2.2. In section 2.3, we introduce several computa-
tional constructs used by the various objects of IRIMAGE. With these fundamentals
23
introduced, we go over each major object of IRIMAGE in some detail in section 2.4.
In addition, this section contains a simple walk through how an image is generated
by the simulation. Finally, in section 2.5 we discuss a a few test simulations using
IRIMAGE and discuss the long-term usefulness of IRIMAGE as both a research and
design tool.
2.2 Physical Basis of IRIMAGE
Several major physical relationships need to be reviewed before a clear discussion of
IRIMAGE can take place. In any infrared simulation one must discuss blackbody
radiation and the effects of the atmosphere. Our simulation employs a ray tracing
method to determine atmospheric profiles and background surface temperatures and
emissivities, thus a brief introduction to ray tracing is also presented. Since we
simulate an imaging system, a clear discussion of the optical collection of radiation and
its conversion to an output voltage from a detector scheme is necessary. Finally, we
discuss the concept of simulating the background fluctuations of radiation that create
the incident photon noise. These ideas form the physical basis for the IRIMAGE
simulation. Most of these concepts have been discussed thoroughly by others; we only
wish to point out the pertinent equations and ideas that directly relate to IRIMAGE.
2.2.1 Blackbody Radiation
Several principles govern the radiation of energy from an object. Kirchhoff’s Law
defines the two basic principles regarding the interaction of electro-magnetic radiation
and objects:
1. All objects may only transmit (7), reflect (p), and absorb (qa) incident radiation
and their three coefficients must sum to one: T+ p+a=1
2. Good absorbers are also good emitters meaning that at thermal equilibrium all
energy absorbed by an object must be radiated.
24
Using Kirchhoff’s law, a blackbody is defined to be a perfect absorber over all
wavelengths and angles (i.e., a = 1,7 = p = 0). Therefore, at thermal equilibrium,
a blackbody must also be a perfect emitter. The Stefan-Boltzman relation further
establishes that the total energy absorbed or emitted by a blackbody only depends
on the temperature of the object. From these principles, the general nature of a black
body is defined. However, none of these properties describe the spectral nature of a
blackbody. Max Planck used the quantized energy states of light and the statistical
nature of quantized energy states to predict the spectral dependence of blackbody
radiation. Planck’s Radiation Formula demonstrates how the radiance, or radiated
power per unit area per solid angle, of a blackbody depends solely on the temperature
of an object (T') and frequency (v) of the light it radiates:
Laplv, Tan = ——” ag 2.2.1
eal, T)dQ = Sa (2.2.1)
From (2.2.1), one can derive Wien’s Displacement law which describes the rela-
tionship between the temperature of a blackbody and the peak wavelength of the
radiance curve:
Amar l' = 2897 um (2.2.2)
Using (2.2.2), we find that for a blackbody at room temperature (T = 293K) the
maximum point of the radiance curve occurs at a wavelength near 10 wm. As the
temperature increases this wavelength decreases. Since most objects on earth have
temperatures in excess of 200 K, one can see that the peak wavelengths will remain
well within the infrared (IR) region of the electro-magnetic spectrum (~ 1 ym — 200
pm). However, most objects are not true black bodies, but are more characteristically
“gray bodies” which can be approximated using the blackbody radiation formula
(2.2.1) and the object’s emissivity (e):
___ Radiance of Object
~ Radiance of a Blackbody
E a (0 25 equation can be defined using the properties of emissivity and equation (2.2.1): Les vy, T)dQ = e(v,T)Lgp(v,T)dQ hy® (2.2.3) In most cases, an infrared imager will be capable of seeing in one of the two main modeling objects. 2.2.2 Modeling the Atmosphere using MODTRAN Various methods have been developed to model the atmosphere. Most of them are HITRAN database [7]. Both models use wave numbers (cm~') as the spectral unit 26 As mentioned, the MODTRAN model is not only capable of computing the spec- atmosphere. Atmospheric Transmission Coefficient (rT) The computation of the atmospheric transmission coefficient uses the MODTRAN is simply the product of the individual gas transmission coefficients for this layer: Talv,l) = [To (¥, Dt. (v, )tc02(v, 1)... |n (2.2.4) Once 7, is computed for each layer along the path, one can compute the total along that path. For a path passing through N layers of an atmosphere, the total 27 transmission coefficient is defined by the following equation: Tatm(V) = If mv) (2.2.5) Background Surface Radiance MODTRAN uses a simple background surface model to calculate the spectral the background surface. Atmospheric Radiance MODTRAN assumes the gas radiance can also be computed using the Planck Ra- radiative transfer function for a single layer [11]: We extend the above case to multiple layers by computing each layer’s individual all of the transmission coefficients for layers between current layer and the imaging 28 system. Env) = Lep(n,v)(1 — talv)) Tn) (2.2.8) where 1 ifn = 0, I17;(v) ifn >0. Then summing over each layer we can get the total contribution of the atmosphere. Les (On.Y)(Tn-1(v) — Tn(v)) (2.2.9) ita Latm( = 2am Total Radiance along the Path TRAN computes the final spectral radiance for any chosen path to be: Lrota(v) = Latm(v) + Leurt(V)Tatm(V) (2.2.10) Despite all of its functionality, MODTRAN does have some drawbacks. Although 29 2.2.3 Determining Optical Path using Ray Tracing Backward ray tracing, considered to be an accepted method for generating physically Various methods may be employed to generate rays [12]. The monte-carlo method In essence, the path method employed by MODTRAN is just a ray tracing calcu- but it allows each detector to sample multiple regions of the background to generate 30 Example of Backwards Ray Trace Monte Carlo Method Background Backwards Array Multiple Rays from A B Projected FPA Projected FPA Nodal Points Nodal Points Single Ray Active Area for Each Submesh element Single Ray 2x2 Submesh Figure 2.1: Backwards ray tracing methods: (a) Example of a backwards ray trace. 3l Although various algorithms, equations and transfer matrices have been developed user to enter the coordinates of these points into the optics parameter file. 2.2.4 Computing the Incident Radiation upon the FPA The ray tracing approach defines a ray’s path through the atmosphere as well as Figure 2.2 shows the generalized situation of an arbitrary surface (S,) emitting is usually not along either surface’s normal, but tends to be at some angle. If we 32 Front of Source of Radiation Figure 2.2: Generalized case of a radiating surface and receiving surface. assume the radiating surface is Lambertian, i.e., radiates equally in all directions, small compared to rj2 and can be neglected. Effective Area of S; : dA; = dS; cos 6; (2.2.11) Effective Area of Sy : dA = dS> cos Og (2.2.12) d dS> cos 0 T12 Tie (2.2.13) Using the above quantities, we can then define the collected spectral power at S» to 33 dS>cos5 (2.2.14) = Lep(v)(dS,cos6,)( ) Tyo The one drawback to the backward ray tracing approach is that the detection Despite the ease of this approach, it adds several more steps to calculating the two solid angles will be equal: AQ detector = A sur face (2.2.15) 34 Nodal Points Detector rojected Area for Solid Angle Surface Solid Angle Figure 2.3: (a) Relationship between detector element area and project surface area. solid angle equivalence. 35 these values in (2.2.15), we get the following: dA le 8 e dAsur O sur Assuming that in the paraxial limit, dsurp < surg and dact X Tact, then Vsur f + A sur f ~ Tsurf Tdet + ddet © Taet By using this approximation and manipulating the terms, equation (2.2.16) becomes Agu ¢ COS Osurp = dAdet COS Oger (—27£)? (2.2.17) Finally, let S, be the radiating surface and S_ be the aperture of the optical system. aperture, Pape, is: dA der C08 One) (dAams C08 Op "det Equation (2.2.18) describes the power collected using the single ray approach. like a smaller detector with its own detection ray. By summing up the power collected 36 for each rectangle, one can get the total power collected for a detector using the sub- Payelv) = So (Papl) i (2.2.19) i,j=0 transmission coefficient and spectral power at that frequency: In IRIMAGE, the transmission curve is defined by a series of data points joined Since we assume that all of the power collected by the optical system is either ab- incorporate the background fluctuation noise. 2.2.5 Modeling Detector Response Once the incident spectral power is determined, it is converted to an output signal by responsivity is usually independent of the photon frequency except near the cutoff 37 responsivity and the total power. Photon Detector: Veig = | RW)P(r)6u (2.2.21) Thermal Detector: Vsig = R | Pw)sy (2.2.22) MODTRAN does not output the spectral power in terms of a function, but returns (2.2.22) to the form of a Riemann sum: Umin tN Av The responsivity of a detector is determined either by using a sophisticated device case. Another method simply treats the data as a piece-wise continuous curve where 38 From this type of representation, the responsivity can be interpolated between by linearly interpolating the given frequency between the two known responsivities. R(v) = M(v — Yn) + R(vn) _ Rn) — RWnt1) (vy — 1) + Rl) (2.2.25) Although this method is less elegant than a fitting method, it makes modeling their own detector routines and incorporate them into IRIMAGE. 2.2.6 Modeling Background Fluctuation Noise Although computer simulations are an excellent way of getting precise answers, pre- Background fluctuation of IR radiation is due to statistically random variations of the radiant flux of photons, number of photons emitted per unit area per second, 39 - Actual Responsivity _ oo oe ae , Interpolated ~ & : Specific Computed \ Spectral Units R(v,)-R Figure 2.4: (a) Responsivity curve composed of specific data points connected by AO upon Ap in time fj, should be: (AN)? =[N—-NP x N (2.2.26) Incorporating the discreet spectral nature of MODTRAN: (AN)? = $7 [N(v) Av — N(v) Av}? = Doan Since both AN? and W are just numbers, any problems with unit dependencies have Our noise model utilizes a known probability function and a random number Given the spectral power, P(v) = N(v)hv, and using the results of section 2.2.5, one can define AVjoise(v) to be: AVnoise(V)Av = (V(v) — V(v))Av = R(v)hvANAv Al Summing over all available frequencies: AV poise = S> AV (v)Av (2.2.29) = © R(m)R(2)h? 4 12AN()AN (1) (Av)? 1,2 We do not average the responsivity because we assume that it is either defined empirically or is already averaged. In addition, AN(v,)AN(v2) can be reduced to For Vy= 1%: AN(@)AN() = (AN()y2 For 1, 4 V2 AN(1))AN (v2) = [N(1) — N()]LN (v2) — N(v)] = N(™m4)N(v2) — N(m)N(2) — N(n)N (2) + N(4) N (v2) since N(v,) and N(v2) are independent variables, 42 Plugging these two cases into (2.2.30) and incorporating (2.2.27) (AVnoise)” = >) (R(v)hv)*(AN(v))?(Av)? Equation (2.2.31) defines the standard deviation of the noise voltage from the function, F(a). Equation (2.2.32) defines F(a) to be the probability of finding a value that is less If we were to reverse this idea and assume we know the probability density we we reduce equation (2.2.32) to an equivalent sum: F(a) = S> AF(z) = S> P(x)Az (2.2.33) EL=—OO L==—OO In figure 2.5a, we show the classic graphical representation of a Reimann sum for 43 x=0 Probability Individual x, Xis1 AF(X,) | AF(X.) I I 1 I Figure 2.5: Number line representation of integrating the Probability Density func- The real advantage of this methodology is that the size of each AF'(x;) region near the peak of the Gaussian corresponds to large range of F values compared to 44 from the following method: e The mean voltage: pu = 0 e The voltage resolution: Ag = 2Aa,,/N 7. Find the noise voltage value, a, by summing over the array of AF (z;) values until F(x) > Fo making a = z. Although the mean value, y, should be the signal voltage, there are several ad- a would have the mean value added to it. Since we are interested in outputing this 45 signal, noise and total voltages, it is just as easy to sum the noise and signal voltages The voltage range and value of N may effect the outcome. The voltage range provides a reasonable value for the resolution. 2.3 Computational Concepts In the previous section, we discussed the physical principles that govern the IRIMAGE discussed here with respect to IRIMAGE. 46 Front View Side View Wy Imaging Optical Axis Front Surface Figure 2.6: Orientation of WCS with respect to the imaging system. The WCS is 2.3.1 The World Coordinate System (WCS) As mentioned, any physical simulation of an imaging process depends on the place- Each simulation element must be defined with respect to the WCS or must be 47 just the inverse of equation (2.3.1). Mecs—wes = R,RyR,T (2.3.1) Once a point in an ECS is transformed to the WCS, it can subsequently be ECS;. Pwos = Mecs,—wcesPacs, (2.3.3) Pros; = Mwes—ecs; Pwos = Mwes—zcs;Mercs,;wesPscs, (2.3.4) These equations demonstrate the usefulness and the need for maintaining a single primary coordinate system like the WCS. 48 Wz Gz Vv bale va 1 aor = iL / WY — ——> Wy Gx Figure 2.7: Relating the grid coordinate system to the world coordinate system. 2.3.2 The Versatile Grid Object (VGO) The Versatile Grid Object defines a multi-dimensional, orthogonal grid structure for Each VGO uses equations (2.3.3) and (2.3.4) to relate coordinates between the The VGO is capable of storing various types of data. Each grid point maintains 49 either a floating point or integer data array. Since the VGO is designed to be flexible, Three of the six major simulation objects of IRIMAGE utilize the VGO. The Scene Since interpolating data onto a grid depends upon the dimension of the grid, interpolated and prepared for loading into a VGO. 50 As mentioned above, the Scene simulation object constructs a background surface Virtual Image Objects The Scene object does not load images directly onto the imaging grid. Instead, Since the design of the Scene object includes animation capabilities, each VIO ignates whether the VIO is active or not. If the VIO is active, then it uses the 51 Original Image Scaled & Rotated Compositing Figure 2.8: Porche image being processed by VIO (scaled, rotated and composited). parameters to process and interpolate the associated image onto the Scene object’s In IRIMAGE, the background images are typically made up of several images that data set needs a matte, which is an image that defines what regions of an image 52 to use in the composite and what regions to discard. An extension of this method An important factor in compositing is the order that images are loaded onto Near Point and Bilinear Interpolation. Each VIO that loads an image serves as the link between the GCS of the Scene’s are not interpolated. In addition, the rectangular nature of the four image points 53 Current State of Image to be Composited Image Matte Interpolated onto ee : Final Composite on Figure 2.9: Example of the matte process. (a) The original Porche image. (b) The background image of a temperature gradiant. (c) The matte of Porche overlayed on 54 In figure 2.10c, a Scene grid point, P, lies inside the rectangle formed by four image reuires more time to compute. Both methods are available in IRIMAGE. 2.3.4 Interpolation of Volumetric Data onto the VGO The Versatile Grid Object is an elegant method for modeling a 2D background surface a) Scene Grid Image in Scene Grid Scene Grid in Image a,b,c,d = Area of Rectangles P= .7 (Nearest Point) P = .2*a+.7*b+.7*c+1.0*d Cc D Figure 2.10: Interpolating images on the Scene object’s imaging VGO. (a) Projecting 56 For most cases, these Ray-Grid methods are only concerned with finding the Ray- One can see how important it is to accurately interpolate a GM onto a three di- specifically designed for interpolating bounded surfaces onto a three-dimensional or- 57 thogonal grid. However, the UPO has several limitations. It only defines the surface When a GM is ready to be interpolated onto a VGO, the GridObj utilizes a “slice a three-dimensional VGO. 2.3.5 The Ray Object The Ray program object generated by IRIMAGE is designed to interact with the 58 In addition to the geometric properties, the Ray also stores a database of grid to generate the atmospheric profile. 2.4 Major Elements of IRIMAGE Now that the physical and computational concepts have been introduced, we briefly the entire simulation process and discuss two test cases. 2.4.1 The Scene Object The Scene Object uses a two-dimensional imaging “plane” to define the background IRIMAGE. However, our initial research interests coupled with the general need for 59 The Scene simulation object models the background imaging scene by compositing The Scene object maintains a database of image data, storing each image in a The processing, placement and layout order of the loaded images are controlled ture and emissivity image data is composited with the imaging VGO’s corresponding 60 The Scene object intercepts each Ray object with the imaging VGO to determine in computation beneficial. 2.4.2 The Atmosphere Object The Atmosphere simulation object defines the atmospheric conditions for IRIMAGE In the Atmosphere object, the VGO plays an important role in defining the at- mospheric conditions. Like the Scene object, the user supplies the origin, orientation, 61 size and grid point spacing of the three-dimensional VGO. Each grid point stores a The Geometric Model Database program object stores and manipulates several the Layout object. For each frame of the simulation, the GMD uses the Layout object 62 to select which GMs to interpolate onto the Atmosphere VGO. Each GM is created in After the geometric models are interpolated, the Ray-Grid intersection algorithm Since the Ray object stores the grid point indexes and path lengths in the order that each grid box was struct, the Atmosphere object can build the atmospheric 63 Active Grid Region | Ys Ya Ps Bounding —_— Figure 2.11: 2D representation of VGO construction using planes. profile using this stack of parameters. The process of building a profile consists of The Ray-Grid interception technique does not store any atmospheric model in- ceptions once which significantly cuts down on the number of computations between 64 Scene Atmosphere Primary Elements of Step 1: Ray exits Optical System Compute distance between A B Figure 2.12: Two-dimensional representation of a ray intercepting a three-dimensional atmospheric VGO. 65 The MODTRAN program object defines the data structures and routines that set Like the Scene Object, the Atmosphere Object is designed to easily incorporate of the Atmosphere simulation object. 2.4.3 The Focal Plane Array (FPA) Object The Focal Plane Array simulation object drives the entire IRIMAGE simulation. where each grid point represents a stack of one or more detectors. Once the VGO is 66 setup, the FPA generates a single Ray object or set of Ray objects for each detector The VGO, combined with a unit cell approach, places detectors in a regular fash- fore, the VGO only stores a single integer index value, which significantly reduces the 67 An 8x8 Focal Plane Array Unit Detection Cell Individual Figure 2.13: Example of an 8x8 FPA using a 2x2 Unit Cell. amount of space required to store a pattern of detectors. In addition to detector layout, the FPA object also initializes and maintains the The FPA object shepherds the Ray objects through the Optics, Atmosphere and Scene simulation objects to generate the spectral radiance. Prior to computing any 68 frames, the FPA passes the rays to the Optics object where the the Ray object At this point, the two approaches briefly diverge again to compute the incident Once the incident spectral power has been computed for a detection element, the file. Using the detection element index, the FPA object accesses the detector types 69 and ID indexes and gives them, along with the element’s incident spectral power, From the discussion above, one can see how the FPA object drives the simulation. any changes. 2.4.4 The Optics Object Like the Scene and Atmosphere simulation objects, the design of the Optics simulation that approximate the spectral filtering of a real optical system. These curves are 70 used to attenuate the spectral radiance returned by each ray after the MODTRAN The Optics simulation object stores several user-defined parameters that define Under the pin-hole approximation, ray tracing through the optical system becomes figure 2.14b, a ray in an optically thick system will not exit the optical system at the 71 Cardinal Points of Optical System n=air n=air Origin of Determining the Exit Point of a Ray Figure 2.14: (a)The cardinal points of an optical system: (i) focal points, (ii) principle 72 the coordinates of the 1% nodal point, Ppt: od Borg = Lerit = Pp + tRair (2.4.1) This equation assumes that within the paraxial limit, t is approximately the same Once MODTRAN calculates the spectral radiance for a ray, the Optics object at- is passed back to the FPA object. 73 Four types of detectors are modeled inside the Detector simulation object: photocon- Before loading the parameters for a specific detector, Detector object must first All four of the current detector models are based on the DetType program object. the DetType object needs the effective area, the spectral range and bandwidth. In 74 Since the background fluctuation model defined by equation (2.2.31) must use the wave number and bandwidth, N(v)Av, for a specific integration time, tine: In addition to the spectral power, the responsivity must also be converted to the signal using equation (2.4.3): v= 3° R(v)N(v)Av (2.4.3) V=Umin Det'Type generates the output noise, V,, using the Background Fluctuation Noise Flexibility is the most important thing to understand about the DetType program the computed output, or even compute the detector output using another algorithm 75 existing ones. 2.4.6 The Output Object Once the detector output is computed, the FPA simulation object passes the resultant structure: In addition, the user can specify whether the incident power and detector output are stored as their actual value or the Log; of their value. This approach allows the 76 2.5 The IRIMAGE Simulation With the introduction of the six simulation objects of IRIMAGE, we can address simulator. 2.5.1 The Graphical User Interface (GUI) In IRIMAGE, all of the settings for each simulation object are loaded from the ob- the parameter files utilize English descriptions making it possible to print out each file 77 Prior to the development of the GUI, the only way to set or change the values GUI provides the user with an effective way of loading parameters into IRIMAGE. 2.5.2 How IRIMAGE Computes An Image Once the user defines the parameters for a simulation, IRIMAGE proceeds to compute parameters have been set, the FPA object creates the layout by assigning a detection 78 The first step in creating an image, IRIMAGE must generate the Ray object(s) Even though each detection element may contain a different detector type or set steps: 79 . Load VIO’s for current frame onto the Scene’s imaging VGO. . Interpolate Geometric Models for current frame onto the Atmosphere VGO. . Open output file for current frame. . Initialize the FPA object’s variables. . Compute the output of each detector using FPA Object. Locate current detection element. Scene object determines background temperature/emissivity for Ray ob- ject. Using MODTRAN, compute the spectral radiance and return it to Atmo- sphere object. Optics object adjusts spectral radiance using the spectral transmission curve. FPA object uses detection element information to convert spectral radiance object. Store or add the spectral power to the total incident power of the detection element. For mesh sizes greater than 1, steps 5d through 5j are repeated until all Ray objects associated with the current detection element are computed. Pass detection element’s total incident power to it’s associated detector(s). 80 (n) Use incident power with the detector responsivity curve and the noise model to generate the output signal and noise. (0) Detector object returns the OutVal data structure which includes the sig- nal, noise and total output. (p) Pass OutVal structure to Output object which stores the output values, incident power and coordinates in the current frame’s output file. As one can see, the process of computing each frame is fairly straightforward. It is be easily incorporated in this procedure. 2.5.3 Two Test Cases In order to judge the effectiveness of the methods employed by IRIMAGE, we con- to address with IRIMAGE in the future. The two cases addressed here are: 1. Simulating passive imaging of a rotating ellipsoid of high concentration, carbon monoxide (CO) at high temperature. 2. Simulating passive imaging of a car moving across a varying temperature back- ground with a cone of carbon monoxide exiting the tail pipe. 81 Parameter Value FPA Size 256 by 256 Detector Type InSb Photovoltaic Mean Responsivity 1.902107" volts/photon Optical F/# 2.545 Table 2.1: Amber IR, camera system parameters for the two test cases. Although a quantitative evaluation of IRIMAGE is necessary and is currently taking Before addressing either case, there are a few general similarities between the two. In addition to the imaging system, both simulations utilize the similar atmospheric at room temperature. 82 Like most gases, CO absorbs and radiates infrared radiation. This property is Case 1: Rotating Ellipsoid of Carbon Monozide. Our first test case examines the ability of IRIMAGE to resolve a localized region In figure 2.16, we present four images from a single frame of this animation. images span a much smaller range from 0.73 to 0.755 volts. In addition, the images 83 Absorbed Radiation Emitted Radiation Rotational Vibrational 10° a U i] T 10° r l l 4.0 5.0 6.0 Figure 2.15: Detection of carbon monoxide (CO): (a) Vibrational and rotational states 84 c D Figure 2.16: Frame 64 of Rotating Ellipsoid of CO in a clean atmosphere. (a) No 85 in (b) and (d) include the noise model contribution, while the other two images This simulation demonstrates the effectiveness of the atmospheric VGO-GMD Case 2: Car Moving Across a Background. This test demonstrates the ability of IRIMAGE to simulate a composited back- bottom half is fixed at 273K. We model the car exhaust with a cone of gas at 323K 86 Cc D Figure 2.17: Car moving across a temperature gradiant background with a cone of and containing 10,000 ppmv of CO (1% CO). This cone translates across the field of Figure 2.17 shows four different frames of the animation. Because the temperature Although it is nearly invisible, these frames do incorporate the noise model. In this 87 case, we can see that the car does clearly stand out against the background image and abilities of both the Scene and Atmosphere objects appear to work well. 2.6 Conclusions The previous examples show just a few of the uses of IRIMAGE. IRIMAGE has the will be best suited for the methodolgy. 88 In general, one can see how the object-oriented approach makes IRIMAGE an We feel that IRIMAGE represents the new philosophy in applied research where research topic before any expensive investment in equipment is decided upon. 89 Bibliography [1] J. R. Schott, R. Raqueno, and C. Salvaggio, “Incorporation of a time-dependent A.D. Sheffer and J.M. Cathcart, “Computer generated IR imagery: a first prin- F.X. Kneizys, E.P. Shettle, L.-W. Abreu, J.H. Chetwynd, G.P. Anderson, W.O. J.L. Meléndez and C.R. Helms, “Process Modeling and Simulation for For more information contact: Dawn Technologies Inc., 491 Macara Avenue, Sunnyvale, CA 94086. A. Berk, L.S. Bernstein, and D.C. Robertson, “MODTRAN: A Moderate Reso- R.A. McClatchey, W.S. Benedict, S.A. Clough, D.E. Burch, R.F. Calfee, K. F.X. Kneizys, L.W. Abreu, G.P. Anderson, J.H. Chetwynd, E.P. Shettle, A. Berk, 90 [9] F.X. Kneizyz, et al., “The MODTRAN 2/3 Report and LOWTRAN 7 Model,” [10] W.L. Godson, “The Computation of Infrared Transmission By Atmospheric Wa- [11] F.X. Kneizyz, et al., “The MODTRAN 2/3 Report and LOWTRAN 7 Model,” [12] An excellent review of Ray Tracing can be found in: A.S. Glassner ed., An [13] W.H. Press, S.A. Teukolsky, W.T. Vetterling, and B.P. Flannery, Numerical [14] A.S. Glassner, “Space Subdivision for Fast Ray Tracing,” JEEE CG&A, 4(10), [15] A. Fujimoto, T. Tanaka, and K. Iwata, “ARTS: Accelerated Ray-Tracing Sys- [16] J. Levine, Programming for Graphics Files in C and C++ (John Wiley & Sons 91 Chapter 3) Multi-Spectral Experiments 3.1 Introduction Any good simulation requires a firm footing in the physics that governs the system In an effort to codify the verification process, we laid out three criteria that IR- are: 1. The Scene object accurately emulates a simple background scene that has re- gions with varying temperatures. 92 effects of a three-dimensional heterogeneous atmosphere. 3. The Imaging system (made up of the FPA, Optics, Detector and Output ob- parameters of the experimental setup. Using these criteria in conjunction with the desire to investigate pollution de- chapter. 93 Since the third criterion requires that the simulated imaging system generates images is done, the images produced by the camera should be comparable to those generated by IRIMAGE. 3.2.1 The Amber AE-256 Imaging System Throughout the experimental verification phase we used a single infrared imaging The Amber camera consists of several key components: the FPA, a filter wheel, liquid nitrogen cooled dewar. A ferro-fluidic coupler minimizes the thermal effects of 94 Temperature Stepper Motor AE-256 Control Box EVC-100 Data Acquisition Video Monitor Figure 3.1: Amber AE-256 infrared imaging system with 5 position filter wheel plus the external filter wheel drive shaft and the internal drive system. A calcium fluoride The support hardware allows the user to calibrate and retrieve image data and the user has the option of setting the integration time, frame rate and initial offset and 95 LN, LHe Output Ge Optics Filter Wheel FPA and Figure 3.2: Drawing of major components of Amber AE-256 infrared camera. gain values for the entire FPA. This is particularly useful in multi-spectral experiments The electronics control box provides the image data in both standard video form black and white or in 12-bit false color. ‘The video signal may also be recorded on any 96 Parameter Value Cint .45 pF Mean Responsivity | 1.90210~' volts/photon CaF Transmission 91 Cold Shield FOV 22.2° Table 3.1: The fixed parameters for the Amber AE-256 camera. VCR directly in black and white and in color if an RGB color encoder is available. For the verification experiments, most of the Amber imaging system parameters used by a particular experiment. These parameters are set prior to the calibration 97 and remain fixed until another filter is selected and the camera is recalibrated. 3.2.2 Simulating the Imaging System using IRIMAGE In addition to modeling the background scene and atmospheric conditions, IRIMAGE Although the major parameters of the imaging system are accounted for, there are some parameters that are ignored. The simulation does not account for the uniformity 98 Parameter Value QE 65 Mean Responsivity | 1.90210‘ volts/photon CaF Transmission 91 Cold Shield FOV 22.2° Table 3.2: The fixed parameters IRIMAGE uses to model the AE-256 camera. and calibration procedures which may set individual pixel gain and offsets. It also IRIMAGE does not generate images directly. Instead, the Output simulation 3.2.3 Comparing Experimental Images with Simulated Im- ages from IRIMAGE. Several important issues regarding data analysis and image comparison need to be discussed briefly. As mentioned above, several parameters for each pixel are set 99 automatically by the calibration procedure of the Amber system; these include the With this idea in mind, we created a method of generating standard images in the For each experiment, we need to define the extreme temperatures of that experi- and 300 and the hot plate is between 3800 and 3900. Staying 5% above the minimum 100 A B Figure 3.3: Varying the range of a captured image to enhance details. Example of and 5% below the maximum provides some tolerance for an experiment which has a When the experiment is over, it is necessary to convert the collected 12-bit im- generated by the verification experiments. 101 The two standard images are very useful in setting the initial range for a set of Once all of this experimental data has been saved, IRIMAGE uses the experiment Once the IRIMAGE simulations are complete, including the images of the hot Provided that the maximum and minimum for an IRIMAGE simulation are set according to the hot and cold plate image, any images generated using this range 102 Figure 3.4: Calibrating the experimental and simulation images of the methane ex- 103 A B Figure 3.5: Extending figure 3.4 to the case of a range enhanced version of the can be directly compared with each other. Figure 3.4 demonstrates how this works IRIMAGE maximum (Viree) and minimum (Vipin) detector voltages. Vinax _ Vinin Figure 3.5 shows the methane case from figure 3.4 with the images enhanced by this method. All of the image data presented in this chapter follows this procedure 104 exceptions. 3.3 Pollution Detection Using Multi-Spectral Imag- The optical detection of any element or molecule in a gaseous state relies on some type Work related to active multi-spectral infrared imaging of the earth’s surface and development of reliable 2D FPA’s, similar systems were designed that use more so- 105 phisticated optics to split the incident IR radiation in to many spectral lines along Our experiments and subsequent simulations demonstrate the effectiveness and detection. 3.3.1 The Vibration and Rotation Modes of a Molecule In quantum mechanics, the concepts of absorption and radiation of electromagnetic radiation not only apply to the bound states of an electron, but also apply to the 106 vibration and rotation states of any molecules that exhibit a natural dipole moment. Since individual molecules are not rigid bodies, the atoms of the molecule are and is a characteristic of the molecule. Using vp, the general equation for the energy 107 Vibrational Mode Axial Rotational Mode Figure 3.6: The vibrational and rotational modes of CO: (a) Vibration along the bond 108 Energy ———» Figure 3.7: The energy potential for a vibrating diatomic molecule. The valley of this states in the SHO region of the potential is: Since the vibration of the molecule results in the oscillation of the dipole moment, equation (3.3.2) for the energy of a photon generated or absorbed, Epis = |AEn|: An important observation is that the selection rule eliminates any energy level it should be possible to detect these transitions using an infrared detector. 109 mind, the angular momentum can be described by the following equation: Using (3.3.3) and the moment of inertia, I, the energy of a rotational state is: _ L? _ r(r+1)h? by = oF = oT (3.3.4) Radiative transitions between rotation states that involve the emission or absorption absorbed photon, En; = |AE,|, and the characteristic frequency of the photon, p;: hh? For most molecules, the value of v, is in the 100 wm to 1 mm range which is include both a change in rotational state and a change in vibrational state. Therefore, 110 molecule. AEn, = AE, + AE, (3.3.7) Since the vibrational transition energy is the dominant energy value, the direction transmission/emission spectrum for any molecule that has a natural dipole moment. 111 3.3.2 Active and Passive Detection of Gaseous Pollutants Since we are interested in both passive and active imaging of gaseous pollutants, it is Active multi-spectral imaging requires a hot source with a known temperature Passive multi-spectral imaging acts almost in reverse of the active case. The all in the incident radiation. 112 Transmission of Methane (CH 3} Figure 3.8: The transmission curves over the 34m to 5m range (2000 cm™! to 3333 113 In our verification experiments, we investigated both the passive and active methods For all three pollutants, we imaged several spectral regions both actively and the results of these experiments in the sections that follow. 3.4 The Relative Temperature Experiments (RTE) In multi-spectral imaging, the desired spectral information must be separated from The RTE uses a background source that consists of a large area at room tem- perature and a smaller isolated region at a higher temperature. The smaller region’s 114 temperature is controllable and is measured using a thermocouple which measures We report the results for three different temperature differences in three different to demonstrate the effectiveness of IRIMAGE and validate its output. 3.4.1 Experimental Setup The RTE consists of a temperature controlled two-dimensional background source and attached to an adjustable current source and the thermocouple connected to a volt 115 8 " Vv 179 cm Figure 3.9: A side view of the experimental setup of the RTE with a front view of meter, we are able to heat the small disk and measure the temperature difference In addition to the RTE background, we also built two aluminum calibration plates. beginning of each experiment and also serve as the standard plates used to calibrate 116 the experimental image data with the simulated image data. 3.4.2 Experimental Procedure Camera Calibration Since the integration time may be different for different filters, it is necessary to must be performed: 1. Rotate the correct filter into place and reinitialize the camera so that a previous calibration does not effect the new calibration. 2. Place the Hot calibration plate in front of the lens about 30 mm from the lens Hot and Cold plates. 3. Begin the automatic calibration procedure and place the Cold and Hot plates sets the individual pixel gain and offsets as well as correcting for bad pixels. 4. After the camera is calibrated, use the Hot and Cold plates to generate another bad pixel map in Amber-View. Using this procedure, we calibrate the camera for a filter and then conduct all of the experiments for that filter before moving on to the next filter. 117 procedure to capture the calibration images: 1. Set the temperature of the Hot plate several degrees above the expected tem- perature of the RTE and keep the cold plate at room temperature. 2. Place the Hot plate 179 cm from the front of the lens, focus the lens and set maximum pixel value between 3800 and 3900. 3. Replace the Hot plate with the Cold plate and set the global offset of the FPA 100 and 200. 4. Repeat steps 2 and 3 until both the Hot plate image has its maximum between between the two images. 5. Capture images of the Hot plate and Cold plate and record the temperatures and gain values for the camera. Setting up RTE and Capturing Images of RTE Source and the initial and final images can be captured in a few simple steps: 1. Place the RTE background at 179 cm from the lens and attach the heater leads to the power supply and the thermocouple leads to the voltmeter. 3. Capture an initial image of the RTE source (AT ~ 0K). 118 ference, AT. Once the images of the RTE background are captured, the calibration image and temperature plate. The entire procedure must be repeated for every filter. 3.4.3 Simulation Setup The measured temperature differences, calibration plate temperatures, filter specifics we did not rotate or translate the imaging VGO except along Z. Parameter AT =10K | AT=5K | AT=1K RTE Tpiate 297.85 K 297.65 K | 297.75 K Table 3.3: Parameters for RTE experiments using the filter N03322-8 In order to define the background scene, the Scene object loads a series of 24-bit images which contain 8-bit maps for the temperature, emissivity and matte values. 119 Figure 3.10: The 8-bit maps that represent the temperature and emissivity for the 8-bit Images Representing RTE Small Disk Temperature Emissivity Matte Figure 3.11: The 8-bit maps that represent the small heated disk. 120 Parameter AT =10K | AT=5K | AT=1K RTE Thiate 297.75 K 297.55 K | 297.55 K Table 3.4: Parameters for RTE experiments using the filter N03990-4 Parameter AT =10K | AT=5K | AT=1K RTE Tyiate 297.15 K 297.35 K | 297.85 K Table 3.5: Parameters for RTE experiments using an open filter position (CaF, win- These maps are converted from the 8-bit values into the corresponding temperature, In all of the IRIMAGE verification simulations, we assumed that the temperature values specified in the provided tables. Although we fix the temperature for each 121 to the experimental conditions and the subsequent images collected by AmberView. 3.4.4 Comparison of Image Results In this section, we show four figures that compare the experimental results of the RTE In figures 3.12 through 3.14, we see how the three filter ranges image the RTE The results are similar for the 5K difference between the large plate and small disk narrow bandwidth (v = .071um) and peaks near 3.3 yum, indicating that the detector 122 E F Figure 3.12: Images for the AT’ = 10K case of the RTE. a) Narrow filter (N03322-8, Pixel Range = 0 to 3800). b) Narrow filter (N03322-8, Voltage Range = 39mV to 407mV). d) Wide filter (N03990-4, Voltage Range = 310mV to 1.37V). ) Open filter (none, Pixel Range = 0 to 3700). Open filter (none, Voltage Range = .72V to 2.47V). (e 123 E F igure 3.13: mages for the AT = 5K case of the RTE. 124 For the case of AT’ = 1K, figure 3.14 shows that the divergence in the narrow filter In figure 3.15, we enhanced the six images of AT = 1K further. This enhancement In general, the images generated by IRIMAGE do compare reasonably well with transmission curves are only approxmations while the responsivity curve, assumed 125 E F Figure 3.14: Images for the AT’ = 1K case of the RTE. (a) Narrow filter (N03322-8, Pixel Range = 3400 to 4000). (b) Narrow filter (N03322-8, Voltage Range = 52.5mV to 97.5mV). (d) Wide filter (N03990-4, Voltage Range = 310mV to 425mV). (e) Open filter (none, Pixel Range = 500 to 4000). (f) Open filter (none, Voltage Range = 815mV to 985mV). 126 Figure 3.15: Enhanced images for the AT = 1K case of the RTE. (b) Narrow filter (N03322-8, Voltage Range = 60mV to 75mV). (d) Wide filter (N03990-4, Voltage Range = 310mV to 334.9mV). (f) Open filter (none, Voltage Range = 815mV to 839mV). 127 to be flat, may have some variance as well. Besides the image comparison, the volt- simulated images with experiment works well. 3.5 The Methane Experiments Hydrocarbons are a common pollutant in the atmosphere. Experimental infrared The methane experiment explores IRIMAGE’s abilities to represent a small but the background using a background image that is a composite of one of the calibration 128 plate images and a scaled and translated version of the same image that represents We report the results of three separate filter experiments and their subsequent possible scenarios for the temperature of the natural gas and the background plate: e A Hot plume against the Hot background plate. e A Cold plume against the Hot background plate. The simulations generate calibration images which are based on the experimental images that can be compared to the experimental results for each filter. 3.5.1 Experimental Setup As figure 3.18 shows, the experimental setup of the methane experiment is very The natural gas plume exits from the end of a 1/4 inch copper tube that is approximately 25 feet long and is coiled up into a 1 ft in diameter coil. This coil lies 129 Methane Experiment AE-256 Plate 202cm Water Bath Vv 243 cm Figure 3.16: A side view of the experimental setup of the Methane experiment. in a plastic tub filled with water. The temperature of the exhaust gas is regulated The natural gas comes directly from the laboratory gas line. Table 3.6 shows the this table was provided by the Gas Company. Molecule Mole Percent Table 3.6: Consituents of natural gas. 130 their contribution. 3.5.2 Experimental Procedure Calibration procedures The camera calibration procedure is exactly the same as in the RTE experiment We start out each filter by measuring a hot plume against a hot and cold back- images of these two cases. 1. Place the Hot background plate 243 cm from the lens and centered on the optical axis. Measure its temperature. 3. Place water bath and coil at position where the end of the exhaust tube is 202 cm from the lens and is 8cm directly below the optical axis. 4. Capture the initial image of the background prior to flowing natural gas through the coil. 5. Open gas valve and let gas flow through the coil. After 10 seconds, capture a sequence of 30 frames. 131 of the tip of the exhaust tube. 8. Replace the Hot background plate with the Cold plate (still 243 cm from lens) and measure the plate temperature. 9. Replace the water in the bath with more hot water and reposition the coil to same position as step 3. Setting up Cold Plume Cases and Capturing images Once we have measured the hot plume cases for the current filter, we repeat the the same. 3.5.3 Simulation Setup Like the RTE experiment, tables 3.7 through 3.9 contain the various temperature the right location and are able to determine the position and location of the burner 132 for the gas plume model. Parameter Hot/Hot | Cold/Hot | Cold/Cold | Hot/Cold T source 329.15 K 298.35 K 298.65 K 329.25 K Table 3.7: Parameters for methane experiments using the filter N03322-8 Parameter Hot/Hot | Cold/Hot | Cold/Cold | Hot/Cold T source 327.55 K 298.05 K 297.25 K 327.55 K Table 3.8: Parameters for methane experiments using the filter N03990-4 The gaseous plume of methane is represented using two static cylindrical objects. rough approximation of a gas becoming less dense as it moves further away from the 133 Parameter Hot/Hot | Cold/Hot | Cold/Cold | Hot/Cold T source 326.85 K | 297.80 K 297.15 K 326.85 K Table 3.9: Parameters for methane experiments using the filter N04235-4 end of the coil. Both cylinders are 25 cm long and are rotated -90 degrees around The background scene is modeled using the calibration image in 3.10. In this case, experimental data. 134 100% 25% Natural Gas | 75% Clean Air Figure 3.17: Example of how the two cylinders are combined to model the static 3.5.4 Comparison of Results Since the experiments examined the four cases of the methane plume against a back- The images presented are enhanced using the technique described in section 3.2. is probably due to minor errors in the measurement of the calibration plate temper- 135 atures and in the measurement of the gas and background temperatures. For the In figure 3.18 we see the hot plume of methane against a hotter background. It In figure 3.19, the hot plume remains, but the background is replaced with a case, we can see both in the simulation and experimentally, that the methane and 136 Figure 3.18: Images for the hot plume of methane against a hot background. (b) CH, filter (N03322-8, Voltage Range = 467mV to 555mV). (c) Clean filter (N03990-4, Pixel Range = 2900 to 3400). (d) Clean filter (N03990-4, Voltage Range = 1.3V to 1.45V (offset +150my)). (f) COs filter (N04235-4, Voltage Range = 1.327V to 1.347V (diff. image)). 137 Figure 3.19: Images for the hot plume of methane against a cold background. (b) CH, filter (N03322-8, Voltage Range = 132mV to 161mV (offset. +25mv)). (d) Clean filter (N03990-4, Voltage Range = 460mV to 490mV). (e) COz filter (N04235-4, Pixel Range = 500 to 600). (f) COz filter (N04235-4, Voltage Range = 701.2mV to 709.1mV). 138 carbon dioxide filters can isolate and detect a hot plume of each of these gases. The Figures 3.20 and 3.21 also demonstrate that IRIMAGE is capable of reproducing In all four cases, RIMAGE demonstrates that it is capable of accurately modeling 139 Figure 3.20: Images for the cold plume of methane against a cold background. (b) CH, filter (N03322-8, Voltage Range = 103mV to 118mV (offset +30mv)). (d) Clean filter (N03990-4, Voltage Range = 460mV to 490mV). (e) COz filter (N04235-4, Pixel Range = 500 to 600). (f) COs filter (N04235-4, Voltage Range = 701mV to 709mV). 140 Figure 3.21: Images for the cold plume of methane against a hot background. (b) CH, filter (N03322-8, Voltage Range = 467mV to 553mV). (c) Clean filter (N03990-4, Pixel Range = 3650 to 3850). (d) Clean filter (N03990-4, Voltage Range = 1.4V to 1.46V (offset +25mv)). (f) COz filter (N04235-4, Voltage Range = 1.327V to 1.347V (diff. image)). 141 a set of atmospheric conditions. It is also clear that the noise modeling and detector should be explored further both experimentally and by computer simulation. 3.6 The Gas Cell Experiments Two of the primary pollutants in automobile exhaust are carbon monoxide (CO) the results from an experiment. 142 14 Figure 3.22: Relationship between CO and COgz in the exhaust of internal combustion Because of the hazardous nature of these experiments, we designed and built a Using three narrow filters, we isolated the CO and the CO, bands as well as a 143 Input Gas Flowmeter 246cm Vv 283 cm Figure 3.23: The Gas Cell experimental setup. realtively clean band (not in CO or CO, band). In the gas cell experiments, we investigated three plume types using these three filters: e A plume of 10% CO and 90% No. Similar to the methane experiments, we imaged a cold plume against a hot background images. 3.6.1 Experimental Setup As shown in figure 3.23, the experimental setup for the GCE is similar to the methane 12”x12” Cold plate or the 12”x12” Hot plate. The size of the gas cell and the desire to 144 18a Mioxiglass Window “ : Gas Cell in Figure 3.24: Drawing of gas cell design. minimize certain depth of field problems made it necessary to place the source plate The gas cell provides a controllable atmosphere that allows us to investigate haz- cell as well as the gas plume temperature. 145 To Gas Cell To Exhaust a : : Flowmeter/ Figure 3.25: Schematic of gas mixing setup. The CO, CO, and N2 are mixed using a standard gas mixing manifold. This In order to avoid reflections of the camera and the interior of the cell from the 10% COz plume through the windows. Since the methane experiment demonstrated 146 that a plume would be visible for concentrations above .2appeared in the images. For Although the windows do introduce unwanted reflections, we constructed a baffle of sources. 3.6.2 Experimental Procedure Calibration Procedures The camera calibration procedure is exactly the same as in the previous two Again, the image calibration was very similar to the methane case. Once the 147 camera was calibrated for a filter, we captured the hot and cold calibration images. Prior to any experiment, the gas cell is purged of all of the pollutants introduced in the previous experiment. The following procedure describes this purging process. Once a purge cycle is complete, the next experiment can be conducted. For each filter and polluted plume, the procedure for generating an image of a hot we setup the experiment and captured the standard and plume images for this case. 2. Place the Cold background plate at 283 cm from the lens and centered on the optical axis. Measure its temperature. 10. 11. 12. 13. 14, 148 Measure ambient temperature inside the cell. Capture 15 frames for use as the standard in the image difference computation. Fill the water bath with hot water which is approximately 320K. Apply 100mA of power to heater at end of copper tube. Open the direct exhaust valve and turn off the high flow Np» valve. Set the flow rates of No, CO and CO, to match desired concentration percent- ages with gas flowing out the direct exhaust tube. Once the flows have stabilized, open the valve to the gas cell and close the direct exhaust valve. Capture 30 frames of gas plume. Measure the plume temperature with the gas thermocouple. Close the valve to the gas cell and close CO and COz cylinder valves. Begin purge process. Imaging a Cold Plume Against a Hot Background Once all of the images of the hot plumes against a cold background have been captured for the current filter, we repeat the same procedure for cold versions of the same plumes against a hot background. For these plumes, the water must be at 290K and the heater at the end of the tube must be off. In addition, we replace the cold background with the hot background. Other than these changes, the procedure is the same as the hot plume against a cold background. 149 Like the methane experiment, tables 3.10 through 3.12 contain the camera settings, ignored its presence in the simulations because we were unable to acquire its surface temperature. Parameter 10% COy | 1% CO | 6.6% CO, 1% CO Table 3.10: Parameters for CO and CO: experiments using the filter N04235-4. IRIMAGE represents a static approximation of the gaseous plumes of CO, CO» or concentration. The physical dimensions of the two cylinders vary slightly from the 150 Parameter 10% CO, |1% CO | 6.6% CO., 1% CO Table 3.11: Parameters for CO and CO, experiments using the filter N03689-4H. Parameter 10% CO,| 1% CO | 6.6% CO., 1% CO Table 3.12: Parameters for CO and CO, experiments using the filter N04693-4. methane experiment. The small cylinder is 25cm long with the end closest to the The background scene is modeled just using the standard image used for the 151 IRIMAGE is unable to simulate the problems related to the thin plastic windows. mind during the image comparisons that follow in the next section. 3.6.4 Comparison of Results The primary goal of these experiments was to determine the effectiveness of detecting As mentioned above, the windows of the gas cell created several problems with ages very difficult because IRIMAGE was unable to simulate these problems. The 152 Figure 3.26: Images of a cold plume of CO, against a hot background looking through (a) Expt: Full Range = 0 to 3800. (b) Sim: Full Range = 372.5mV to 740mV. (c) Expt: Nar. Range = 2800 to 3800. (d) Sim: Nar. Range = 643.3mV to 740mV. 153 technique of enhancing the captured images by reducing the size of the range and its 154 Figure 3.27: Images of a cold plume of 10% COz against a hot background. b) Clean filter (N03689-4H, Voltage Range = 481mV to 494.1mV). c) CO» filter (N04235-4, Pixel Range = -25 to 175) d) COz filter (N04235-4, Voltage Range = 689.2mV to 708.6mV). e) CO filter (N04693-4, Pixel Range = -75 to 75). (a 155 Figure 3.28: Images of a cold plume of 1% CO against a hot background. (b) Clean filter (N03689-4H, Voltage Range = 483.1mV to 496.3mV). (c) COs filter (N04235-4, Pixel Range = 70 to 170). (d) COz filter (N04235-4, Voltage Range = 705mV to 714.4mV). (e) CO filter (N04693-4, Pixel Range = -25 to 75). (f) CO filter (N04693-4, Voltage Range = 1.508V to 1.534V). 156 Figure 3.29: Images of a cold plume of 6.67% CO2/1% CO against a hot background. b) Clean filter (N03689-4H, Voltage Range = 485.3mV to 498.4mV). c) COz filter (N04235-4, Pixel Range = -225 to -25). d) COz filter (N04235-4, Voltage Range = 677.8mV to 697.3mV). e) CO filter (N04693-4, Pixel Range = -200 to 0). (a 157 important result here is that the experimental results seem to indicate that the CO For the case of a plume that may be a mixture of both gases, it should be possible Figures 3.30 through 3.32 present the imaging results for the other case; a hot the CO filter; these should also be ignored. In figure 3.30, the plume consists of a 158 Figure 3.30: Images of a hot plume of 10% CO: against a cold background. (b) Clean filter (N03689-4H, Voltage Range = 180.3mV to 189.1mV). (c) COs» filter (N04235-4, Pixel Range = -10 to 35). (d) CO, filter (N04235-4, Voltage Range = 409.3mV to 413.6mV). (e) CO filter (N04693-4, Pixel Range = 0 to 100). (f) CO filter (N04693-4, Voltage Range = 670.5mV to 696mV). 159 Figure 3.31: Images of a hot plume of 1% CO against a cold background. (b) Clean filter (N03689-4H, Voltage Range = 194.4mV to 211.9mV). (c) COs filter (N04235-4, Pixel Range = -25 to 30). (d) COs filter (N04235-4, Voltage Range = 416.7mV to 421.9mV). (e) CO filter (N04693-4, Pixel Range = 5 to 20). (f) CO filter (N04693-4, Voltage Range = 683.8mV to 687.7mV). 160 Figure 3.32: Images of a hot plume of 6.67% CO2/1% CO against a cold background. (b) Clean filter (N03689-4H, Voltage Range = 190mV to 203.1mV). (c) COs filter (N04235-4, Pixel Range = -15 to 25). (d) COz filter (N04235-4, Voltage Range = 418mV to 421.9mV). (e) CO filter (N04693-4, Pixel Range = 0 to 20). (f) CO filter (N04693-4, Voltage Range = 688.9mV to 694.2mV). 161 mixture of 10% CO 2 and 90% nitrogen. The IRIMAGE static cone model predicts As before, we investgated the case of a hot plume containing a mixture of nitrogen, predicting the outcome of an experiment. 162 These experiments lead to several conclusions regarding passive, multi-spectral in- It was important to prove that IRIMAGE could generate simulated images that The noise model was another major problem with the simulated images. For 163 the filters greater than 4 microns, the noise model seemed to work quite well. As The methane and gas cell experiments conclusively demonstrated the promise of Despite the success of these experiments, certain limitations do exist. The need to not require recalibration after a new filter is rotated into view. In this case, the 164 IRIMAGE provides the basic tools to make this determination. 165 Bibliography [1] Amber-View Software Manual, Release 1.0.a (Amber Engineering, Goleta, CA, Implementation of the Flexible Image Transport System (FITS) NOST 100-0.3b, AVS is a trademark of Advanced Visual Systems Inc. Inc., 300 Fifth Ave., Waltham, MA 02154. For an excellent discussion of terrestrial monitoring using imaging spectroscopy: G. Vane, R.O. Green, T.G. Chrien, H.T. Enmark, E.G. Hansen, and W.M. A.F.H. Goetz and Mark Herring, “The High Resolution Spectrometer (HIRIS) An excellent review of vibrational/rotational modes is in: R.M. Eisberg and R. Resnick, Quantum physics of atoms, molecules, solids, nulcei, and particles, (John Wiley $ Sons, New York, 1974). Curves based on data from: C.J. Pouchert, The Aldrich library of infrared spectra, J.H. Seinfeld, Atmospheric Chemistry and Physics of Air Pollution 166 [11] Linde Gases of Southern California, Specialty Gases & Equipment Catalog, 167 Appendix A Block Diagrams of Part A of this appendix contains a series of block diagrams of IRIMAGE. These listing of the five diagrams: e Diagram A.1 shows how the six major elements of IRIMAGE are connected together. e Diagram A.2 shows how the Image Database, the VIO Database, the Layout e Diagram A.3 shows how the MODTRAN object, the Geometric Model Database e Diagram A.4 shows how the four detector databases are contained within the Detector Object. e Diagram A.5 shows how the Optics Object, the Output Object, Element Database These diagrams designate the major objects and databases as rectangles; databases demonstrate how IRIMAGE works, only how it is put together. 168 IRIMAGE Object Object Optics Output Optics Output Figure A.1: Diagram showing how the six major objects are connected in IRIMAGE. 169 Scene Figure A.2: Diagram showing the supplementary objects in the Scene Object. 170 Atmosphere MODTRAN Gas ‘A Geometric Atmosphere Figure A.3: Diagram showing the supplementary objects in the Atmosphere Object. 171 Detector Photoconductor |. Photovoltaic Database Pyrometer Database Bolometer Database Database Detector Detector Figure A.4: Diagram showing the supplementary objects in the Detector Object. 172 FPA Element Element Element Ray Radiance Ray Radiance Figure A.5: Diagram showing the supplementary objects in the FPA Object. 173 Appendix B_ Interpolating Volumetric B.1 Introduction There are several approaches to defining a three-dimensional scene in a computer A scene composed of volumetric representations may not give a completely ac- building blocks. When they are stacked in a particular order, they form a volumetric 174 Our approach builds on the basic SOE representation. In most cases, only the In order to effectively use the VGO to represent an entire scene, we developed In section B.4, we discuss the “Slice and Dice” algorithm for interpolating UPO- 175 curves are then interpolated with a two-dimensional grid using a published interpola- In section B.5, this chapter introduces how the VGO, UPO and the interpolation the interpolation algorithm that links them together. B.2 The Versatile Grid Object (VGO) The Versatile Grid Object is a self-contained program object that stores either integer volumes. B.2.1 The Grid Coordinate System (GCS) The VGO has its own internal coordinate system called the Grid Coordinate Sys- 176 Wz Gz Wz Gz Figure B.1: Various VGO transformations: (a) Original GCS to WCS relationship. 177 symmetric octant (quadrant in 2-D) structure. Since most of the applications that The following equations define the basis matrices for the corresponding trans- dimensional computer graphics [7]. The Translation Matrix ((T(d)) and Scale Matrix (S(s)) D0 0 dy, fs, 0 0 0 000d, 0 0 s, 0 lo 00 1] fo 0 0 1 178 1 0 0 oO lcosw 0 sinw 01 Using simple matrix multiplication, we can generate the transformation and in- and care must be taken to define the angles properly. With this in mind, equations (A.2.1) and (A.2.2) define this order. Mwces—ccs = R,:Ry-Rx-T (A.2.1) Changing the grid point spacing vector (G) conforms to a simple matrix-vector multiplication of the scaling matrix, S, and G: Equation (A.2.3) transforms the grid point spacing for all three axes of the VGO. every time a change of scale occurs, it is an absolute change of scale and not relative 179 to the previous scaling. B.2.2 Data Storage in the VGO The internal structure of the VGO is made of a one-dimensional, integer-based Object may store a link to an external data structure which eliminates the need for the other data array. r (tg) = (0,0) ¥ v Ay i=0 i=N i= 2N & od a oo _— & a & a ae 3 cs 8 4 | 4 2 © v a, & sy & — an a a. = > > > c % a= en > > > A 2D Array Representation B Actual 1D Array Figure B.2: Example of how data array is stored in the VGO: (a) Perception of data The FDA can store multiple variable values in each grid point of the VGO. For N-M values. Figure B.2 demonstrates how the data is stored as a one-dimensional 180 array although its access may appear more two-dimensional, i.e., the coordinates and Although the OIA and FDA are one-dimensional in nature, their association with Object Index Array (OIA) & Flexible Data Array (FDA) Indexes OI Aina(X,Y,Z) =X +(Nx)Y + (NxNy)Z (A.2.4) FDAina( X,Y, Z,V) = Ol Aina X,Y, Z) + (Ny Ny Nz)V (A.2.5) 181 where Nx, Ny, Nz = Total # of grid points along each axis V = Index that indicates which element of the FDA should be set or retrieved The VGO may use the integer index of a grid point to calculate its physical point spacing vector, G, the VGO can compute (z, y, z): (A.2.6) Similarly, the nearest integer index can be found by reversing (A.2.5) and rounding Using the WCS to GCS transformation matrices, it is even possible to retrieve the The VGO does not contain routines which interpolate any type of image, UPO algorithms. 182 on the VGO With a VGO in place, it is possible to discuss how one goes about defining a bounded it also stores the volumetric data to be interpolated on the VGO. B.3.1 Defining Bounded Surfaces Using Polygonal Surface In the process of defining the algorithm for intercepting a plane with a bounding most straightforward and popular method, we chose to use it as our basis for inter- 183 In standard CG, polygon surface modeling can be defined in two ways. The In the edge-based method, each polygon stores an ordered list of edges which In a polygon-based surface model, each polygon is independent so that any surface- The Point Array Structure This structure simply stores the three-dimensional coordinates for every vertex The Edge Array Structure Like the edge-based approach, this array stores the pointers (integer array indexes) to the end points of each edge line. Since the model will be intercepted by a cutting plane, we also incorporate two more integer pointers which store the two polygons 184 Two Polygoue Edge Pairs of Edge Pairs of Polygon A Polygon B #1 #2. #1 #2. 1 2 4 5 2 3 5 6 3 1 6 4 Polygon-Based Surface Representation >< Part of a Polygon-based The Common Edge Points are Two Polygons Edge Point Point a1 | 2 Polygon Array {Edge ID) Edge-Based Surface Representation Part of an Edge-based Now both polygons store the Common Edge. Figure B.3: Two methods of defining a bounded surface. (a) In a polygon-based system each edge that is common to two polygons is stored twice (once by each stored only once since polygons only store pointers to edges. 185 that share the common edge. The usefulness of this additional information will be The polygon array structure stores the pointers (integer array indexes) to the since 3 points can only define a plane while more can describe a non-planar surface. B.3.2 The Universal Primitive Object (UPO) Model In the process of developing the VGO-Bounded Surface interpolation algorithm, the The UPO model utilizes two basic constructs, multiple control contours and two the GCS it must maintain the same base unit (e.g., feet, meters, etc.) so that points 186 These contours circumscribe the outer surface of the object. Every contour has the Once the individual contours are set up, they can be connected together or to one Adjacent contours can be connected in two ways. The first method simply con- follows: Method 1 Given that each contour has N points. For contour R,, each Point P; is Pj41 on contour Ry y1. For the case of Pjzn4i, use point P,. The problem with the first method is that it is unable to create parallelepiped a point is displaced, the effect on the surface depends greatly on how it is connected 187 2nd End Point 6 6 Figure B.4: Example of normal connection method for creating an Ellipsoid: (a) The 188 2nd End Point Connected Control Control Contour o~ A 1st End Point |B Cc D Figure B.5: Example of alternating method of connecting contours for creating a 189 models from this method is: Method 2 Each contour has N points. The points along contour Ry are unconnected connected to P; on Rn41. For the case of Pjan41, use point P,. End Points In addition to control contours, the UPO has two end connection points. These Parameters used by UPO to build any primitive object: 1. Pointer to Building Contour Parameters (stored in order). 190 Genter location: C(n) Scale (before orientation): G(n) — Center orientation: R(n) — Use to generate the Array of Points used to build object. ~ Array of predefined points. — This array is simply copied to final object. e Create Point Array (Array Size = NyingNpt + 2). e Create Polygon Array (Array Size = 2NpingNpt). (a) Contours Built by UPO e Load first end point. e Build successive contours beginning with contour closest to first end — Use 2D scaling matrix to scale contour: P’(i) = Sy - P(i). 191 ~ Rotate contour about its own origin: P”(i) = Ry - P’(2). ~ Translate contour to center point in POCS and store point: — Repeat for each contour. e Load second end point. e Load first end point. e Load second end point. e Build edges from first end point to first contour’s coordinates. e Associate edges with cap polygons. (a) Method 1 - All contours are connected. e Build diagonal edges from first contour to second contour. e Build polygons using adjacent diagonals and edge of first contour. Build edges for second contour. Build polygons using adjacent diagonals and edge of second contour. Associate edges with polygons. — Second contour: first polygon defined now. e Build diagonal edges from connected contour to unconnected contour. 192 e Build vertical edges between connected ring and next connected ring. e Build polygons with two adjacent diagonal edges and the edge of first connected contour. e Build diagonal edges from unconnected contour to second connected contour. e Build third set of polygons with two adjacent diagonal edges and sec- ond contour edge. e Build next set of contours if necessary. e Build edges from last contour to end point. e Associate edges with polygons. In order to define various point symmetric or axial symmetric objects, the user The addition of storing polygon information in the edge structure requires no special procedure. One simply adds the step of linking the polygon to the edge right 193 Parameter Boz Ellipsoid Nring 3 3 or more Not 4 >= Nring Build Contours Yes Yes Alternate Contours Yes No End Point Locations Center of R; & R3 Use Equation Table B.1: Parameters for building Ellipsoid and Box UPO after linking the edge to polygon. In addition, the fact that only the point array stores Once the geometric model is defined, the volumetric data is associated with a advantages of the UPO are seen when interpolating UPO volumes onto the Grid. B.4 Interpolating a UPO onto a VGO As mentioned earlier, the “slice and dice” methodology for interpolating a bounding series of voxels which have a different set of characteristics from the rest of the voxels, 194 In general, the characteristics stored in each voxel might include information such is “drilled” or “traced.” Such a situation is discussed in section B.5. B.4.1 Intercepting a Cutting Plane with the Surface of a The “slice” portion of the algorithm requires a regularly spaced set of cutting planes When intercepting a UPO-based surface with a cutting plane, one can see the in generating the intercepted curve, it is extremely difficult to do because the order 195 A Sample Grid Object B Bounded Surface of a Cone Cone Inside Grid Object D Cone Interpolated on Grid Cc Figure B.6: Example of a UPO interpolated on three-dimensional VGO. (a) A wire- 196 Cutting Planes Cutting Planes along Gz Axis Figure B.7: Cutting planes going along Z axis of the VGO. of the intercepted points is variable. Using a UPO-based surface eliminates these The UPO reduces a difficult problem to a simple methodology. Once the UPO is When the UPO-based surface is sliced by a cutting plane, the algorithm cycles coordinates of the interception point where each edge and plane coincide are stored 197 The algorithm begins with the first point in the array of intercepted points. It more closed curves representing the cutting plane-UPO surface interception. B.4.2 Ordering the Intercepted Closed Curves As mentioned, an alogrithm for intercepting a closed curve with a two-dimensional “dicing” algorithm can feed the curves in the proper order to the well-defined curve- 198 the following geometric property: Given: A 2D plane that contains a line segment and a series of non-intersecting, prece-wise continuous, closed curves. Result: One can determine the number of curves that contain all or part of the line segment. Method: Extend a ray to infinity along one of the two directions of the line segment. the line segment. Figure B.10 graphically demonstrates this methodology. Given that all of the curves that come from the bounded surface-plane interception the inner most ones (level N). B.4.3. Using Closed Curves to Store Data on VGO Once the curves are sorted, the “dicing” portion of the algorithm begins. The al- gorithm we use for intercepting a piece-wise continuous, closed curve with a two- 199 Top View of Cutting Plane Level O Top View of Cutting Plane Level O oN. Figure B.8: Example of two types of multiple curve interpolation. (a) Cutting plane B ae Figure B.9: Two examples of a Torus interpolation. (a) Bounded surface model of 201 Extended Ray from Curve D Line Segment inside Extend Ray to determine Extend case to an entire curve C) Final Results re A |Y(1) B |Y(1) . Method can be used Figure B.10: Method for determining how many curves contain a line segment. or 202 Level O Added to Grid Need to Subtract Level 1 Curve D Final Interpolation Result Figure B.11: Example of how two curves are interpolated on a 2D grid (Curves from 203 longer needed and can be discarded or returned to the WCS for use again. B.4.4 The Slice and Dice Algorithm 1. Transform UPO Surface into Grid Coordinate System (GCS). (a) Create Geometric Model as a UPO and transform it to WCS. (b) Transform point array of UPO into GCS. a) Cycle through point array to get absolute minimum and maximum coor- dinates. (b) Use these two points as the basis for the bounding box. 204 3. Use bounding box to determine what cutting planes to use. (a) Use bounding box Z coordinates to define first and last cutting planes. (b) Define a loop to create cutting planes to use in interpolation. 4. Cycle through cutting planes to intercept volume with voxel layer. (a) Create cutting plane object using Z value (i.e., parallel with XY plane). (b) Pass the UPO surface model to Plane program object to determine the interception curves. 1. il. ill. Cycle through edge lines of polygons to find all of the intercepted e Store intercepted point. Cycle through intercepted points to determine all of the curves inter- cepted. Sort curves into levels. (c) Starting with level 0, intercept curves onto the voxel layer. 1. i. ill. Create a temporary two-dimensional grid. e Same grid point placement and size as the original grid in XY plane. Intercept curves onto the VGO using polygon-grid interception rou- tine. e Even numbered levels add to intercepted areas. iv. Use intercepted areas of temporary grid to load the voxel layer of the VGO. A. If at least 50% of the area of a grid rectangle is inside the UPO: 205 e Pass a unique object index to VGO grid point. angle. (d) Repeat process for each cutting plane until entire volume is intercepted. B.5 Application of the VGO and the UPO in the One application of the VGO is modeling a background atmosphere in the infrared The IR system simulator, JRIMAGE, uses a VGO and a set of UPO-based volumes and the FDA to store two integer values for each voxel. Initially, the VGO loads each 206 the incident radiance on a detector system. B.5.1 Generating an Atmospheric Profile in IRIMAGE IRIMAGE utilizes a backward ray tracing approach to compute the atmospheric by accessing each intercepted voxel and retrieving its current state. If the state of 207 As mentioned, the rays store the grid point indexes of the intercepted voxels and Once the rays have been generated and intercepted with the VGO, the various of the UPO for a specific set of frames. All the frames between the key frames are Bounded Volumes Placed Volumetric Data Interpolated Figure B.12: A 2D example of a how an atmospheric profile is generated. (a) Various 209 interpolated linearly. So instead of having to define the conditions for each frame, Once the rays have been intercepted and the UPO’s have been loaded, IRIMAGE well-defined imaging system. B.5.2. Example of Various UPOs Used in IRIMAGE In figure B.13, a polluted region was modeled using four different UPO-based geo- (ppm) or 5% by volume. The standard atmosphere of the VGO had a temperature of 210 Cc D Figure B.13: Various UPO-based bounded volumes representing a polluted region of 211 approximation of a polluted atmosphere’s profile. B.6 Conclusions The object-oriented design of the VGO and UPO makes these objects very flexible. the same object in the same program or another program. For example, IRIMAGE 212 Despite all of its advantages and benefits, the VGO still has some drawbacks. The The algorithm for interpolating UPO-based bounded volumes onto the VGO works combining data sets that overlap in the same voxel. This is all right for solid objects, 213 but inadequate for gaseous and liquid objects which can intermix. Again, the user In general, the VGO, UPO and the interpolation algorithm provide the basis for that is not heavily dependent upon surface representations. 214 Bibliography [1] [10] J. Foley, A. van Dam, S. Feiner, J. Hughes, and R. Phillips. Introduction to Discussion of Constructive Solid Geometry: J. Foley, A. van Dam, S. Feiner, Excellent review of octrees: H. Samet. Design and Analysis of Spatial Data Struc- tures. (Addison-Wesley, Reading, MA, 1990). R. Lohner and P. Parikh, “Three-dimensional grid generation by the advancing front method,” Int. J. Numer. Methods in Fluids, 8, 1988, 1135-1149. P.L. George and E. Serveno, “The Advancing-Front Mesh Generation Method J. Foley, A. van Dam, S. Feiner, J. Hughes, and R. Phillips. Introduction to J. Foley, A. van Dam, S. Feiner, J. Hughes, and R. Phillips. Introduction to D.A.P. Mitchel, “Fast Algorithms for 3D Graphics,” Ph.D. Thesis, University of C. Deutsch, “A FORTRAN 77 Subroutine for Determining the Fractional Area F.X. Kneizys, L.W. Abreu, G.P. Anderson, J.H. Chetwynd, E.P. Shettle, A. Berk, 215 W.O. Gallery, and S.A. Clough, “The MODTRAN 2/3 Report and LOWTRAN [11] J.H. Seinfeld, Atmospheric Chemistry and Physics of Air Pollution, (John Wiley 216 Appendix C Example of Parameter Files This part of the appendix contains an example of the seven parameter files used by the files makes it reasonably easy to determine what each of the parameters are. 217 # IRIMAGE General Parameter File # Simulation Name: Car # Created on Date: 9/3/97 $ IRIMAGE General Parameters Starting Frame Number Output Parameter File = car_sim/script/car.out Num of Imaging Systems= 1 Imaging System Index = 1 Monitor Filename = car _mon 218 # Scene/Image Object (SCN) Parameter File # Simulation Name: Car # Created on Date: 9/3/97 # Scene Loading (scene.cxx) $ Scene Parameters ll Origin of Scene 0.0, 0.0, 5.0 Orientation of Scene 180.0, 0.0, 0.0 Scene Grid Spacing = 2.0e-3, 2.0e-3 Background Temperature= 293.15 Background Emissivity = 0.90 # Image Database Loading (image.cxx) $ Image DB Parameters Number of images = 3 Image File Name bars_t50.ascii Type of Image File = 1 Grid Point:X,Y Spacing= 4.0e-3, 4.0e-3 # Virtual Image Object Pos. of Temperature ASCII:# of Header Line ASCII:X, Y Coords Corner Location Grid Point:X,Y Spacing= Grid Point:X,Y Sizing = Pos. of Temperature Binary:Type of File Binary:Temper. Add Binary:Emissivity Mult= Binary:Emissivity Add Binary:Mask Mult Corner Location Grid Point:X,Y Spacing Pos. of Temperature ASCII:# of Header Line ASCII:X, Y Coords 219 car_bw.pgmb 2.0e-3, 2.0e-3 1.0 0.0 2.0e-3, 2.0e-3 -1 -1 Loading (image.cxx) VIO Parameters Image Index Load Temperature Pos Load Emissivity Pos Load Mask Values Pos It Interpolate using Mask= VIO Object Name Image Index Load Temperature Pos Load Emissivity Pos Load Mask Values Pos Hl Interpolate using Mask= Minimum value of Mask Transparency at Max. VIO Object Name Image Index Load Temperature Pos Load Emissivity Pos Load Mask Values Pos it Interpolate using Mask= 220 Bars_Image 0.0 1.0 100.0 0.0 # Load Animation Parameters (keyfrm.cxx) # Animation Parameters for Each VIO $ Key Frame Param: Bars_Image Number of Key Frames Object Active Origin Location (WCS) # of Interp. Coeff. # of Integer Data Pts Object Active Origin Location (WCS) # of Interp. Coeff. # of Integer Data Pts Object Active Origin Location (WCS) # of Interp. Coeff. # of Integer Data Pts # of FP Data Pts $ Key Frame Param: Car_ Number of Key Frames Orient. about Origin li 221 .0, 0.0, 0.0 rae -O, 1.0, 0.0 oO Oo 8O ~.125, -0.05, 0.0 -0.375, -0.05, 0.0 1.0, 1.0, 0.0 Matte -.125, -0.05, 0.0 222 Scale about Origin = 1.0, 1.0, 0.0 # of Integer Data Pts = 0 # of FP Data Pts = 0 Origin Location (WCS) = -0.375, -0.05, 0.0 # of Interp. Coeff. = 0 # of Integer Data Pts = 0 # of FP Data Pts = 0 # Load Layout Parameters (keyfrm.cxx) # Layout DB Loads VIO Index and Type in Order of Layout $ Layout DB Param: Scene Object Layout Number of Layout Obj = 1 Layout Key Frame Num. = 1 Num of Objects in Fram= 3 Object at Pos #0004 Be Object Type Pos #0001 It ON OD W OO Object at Pos #0001 Object at Pos #0001 223 C.3. The Atmosphere Object Parameter File Simulation Name: Car + + + + HH FH Created on Date: 9/3/97 Atmospheric Object (ATM) Parameter File # General Atm Object Parameters (atm.cxx) $ Atm Class Parameters Origin of Atm Grid = 0.05, 0.0, 2.5 Orientation - Atm Grid= 0.0, 0.0, 0.0 Atm Grid Spacing = Atm Grid Size Use Geom Objects Background Gas Model Background Aerosol Mod= -1 Type of Standard File = 1.0e-3, 1.0e-3, 1.0e-3 200, 200, 150 # Gas Model Database (modtran.cxx) $ Atmosphere: GasDB Model Index: 1 Pressure Value/Unit i] CO Density/Unit oO olUmUDWUlmUCUCOUUCUCOUWUCUNLUN CO2 Density/Unit -60e+02, C CH4 Density/Unit = NO2 Density/Unit = [<> <> © © $02 Density/Unit Pressure Value/Unit = 7.61e+02, C # Aerosol Model Database (modtran.cxx) $ Atmosphere: ArsolDB Num. of Models in DB = 1 Model Index: 1 Aerosol Number Density= 1.0 Equiv. Liq H20 Content= 0.0 Seasonal Modifications= 0 Change Profile Region # Modtran Cards (modcard. $ Modtran: Card 1. Model Type Type of Path Execution Mode Type of Default Scattering H20 Model 03 Model CH4 Model N20 Model CO Model Data Loading Method Oo F Oo Oo 0 0 006CUCU00UCUOUUCUCODD Report Generation Type= Background Temperature= Background Albedo $ Modtran: Card 2. Aerosol Season Aerosol Profile Attenuation Stratospheric Profile Mass character-Navy Cloud Model Type Ml Army Vert. Struct. Alg= Surface visibility km = Wind Speed-Navy/Desert= 24 hr Ave Wind Speed Rain Rate (mm/hr) Surface Altitude cxx) PN 293.150 oo © 2 a © a ©? © >) 225 $ Modtran: Card 2c. Number of Atm Layers User Defined Gas Dens.= User Defined Aerosols = Layer Top Boundry Alt. QO O92 Oo Oo fF O09 90 0 00CU0DUmUmUmUCUDUCCODOUNUUUNCUOULUCODWClCUCOCUCOUCOCUCOTUOCOCODCOUCCOCOTT OCCT a awe NCO Co Layer Pressure/Unit Layer Temperature/Unit= H20 Density/Unit Layer Top Boundry Alt. Layer Pressure/Unit Layer Top Boundry Alt.= [© © © > 2) .0 . ~~ . . . . . . joe a © © > > > .0e-3 ~ . . v . . . “ . « Oo DDD DH DH HD DH DB DBD WD oO 226 Layer Pressure/Unit = HNO3 Density/Unit Layer Top Boundry Alt. Layer Pressure/Unit $ Modtran: Card 3. Earth Center Angle = 227 -60e+02, C . - . . . . “ vy “ [<> > OO > co) . . + . . [a >, > > SO > SP .O0e-2 179.0 ae) 0.0 228 Radius of Earth = 0.0 $ Modtran: Card 4. Initial Frequency WN = 2000 Final Frequency WN = 2300 # Geom Values (geom.cxx) $ Geom Class Parameters Box Type Index = 0 # Box Object Parameters (box.cxx) $ Geom Model: Box DB Number of Box Models = 0 # Cone Object Parameters (cone.cxx) $ Geom Model: Cone DB Number of Cone Models = 1 Cone Model Name = Cone_1 Start. Angle of ist Pt= 0.0 229 Number of Pnts in Base= 16 # Cylinder Object Parameters (cylinder, cxx) # Ellipsoid Object Parameters (ellipse.cxx) $ Geom Model: Ellipse DB Num. of Ellipse Models= 0 # Load Animation Parameters (keyfrm.cxx) # Animation Parameters for Each Object Gas Model # Integer Data Value #1 Aerosol Model # Integer Data Value #2 $ Key Frame Param: Cone_1 Number of Key Frames = 2 Origin Location (WCS) = 0.268, -0.005, 2.5 # of Interp. Coeff. = 0 # of Integer Data Pts = 2 Integer Data Value #01 Object Active = It Origin Location (WCS) Orient. about Origin Scale about Origin # of Interp. Coeff. # of Integer Data Pts Integer Data Value #01 # Load Layout Parameters Hi It 90 149, -0.005, 2.5 (keyfrm.cxx) Index and Type in Order of Layout $ Layout DB Param: Geometry Object Layout Number of Layout Obj = Object at Pos #0001 Object Type Pos #0001 231 Focal Plane Array Object (FPA) Parameter File Simulation Name: Car Created on Date: 9/3/97 ++ + + + H+ FF # FPA Parameters (fpa.cxx) # Load specified index number for FPA $ FPA Parameters # 1 WCS Center Pnt of FPA = 0.0, 0.0, -0.17 Detector Element Pitch= 3.8e-5, 3.8e-5 Num Of Elements (X,Y) 256, 256 Ray Distrib. Method # of Mesh Points (X,Y)= 1, 1 # of Detectors in Cell= 1 Pct Size of Element = 100.0, 100.0 Min Wavenumber to Calc= 2040 # of Lines to Calculat= 100 Full Width - Half Max = 2 232 # Detector Database (DET) Parameter File # Simulation Name: Car # Created on Date: 9/3/97 # General Detector Parameters (detector.cxx) $ Detector Object Number of PV Detect =i Detector Type = 0 Detector ID Index= 1 # Detector Type Information (detector.cxx) # Each detector has this type of loading parameters # (Integration Time = (1/Frame Rate)*(Amber Int Time/128) $ Param-Detector # 1 Det. Cold Shield FOV = 22.2 Fill Factor = 0.954 FPA Output Units Noise Prob Function Curve Index 233 # Photovoltaic Detector Specific Parameters (PV) (detector.cxx) $ PV Detector Param # # Responsivity/D* Curves of Detector(s) (detector.cxx) $ Detector Response Param # 1 Type of Curve Curve Spectral Units ii Measured Noise Voltage= Meas. Cold Shield FOV = Measured Noise Bandwth= Measured Int. Time # of Points on Curve Responsivity/D* Value 0.624e-3 1818.0 234 C.6 The Optics Object Parameter File # Optics Object (OPT) Parameter File Parameter Loading Meth= 0 Front Origin (WCS) = 0.0, 0.0, 0.0 Effective F Number = 2.5489 Number of Trans Curves= 2 Transmission Coeff Ind= i Transmission Coeff Ind= 14 # Transmission Coefficients for Compound Optics (optics.cxx) 239 # CaF Window $ Transmission Coeff # 1 Transmission Coeff. = 0.920 # Filter - NO3288-8 (Cen = 3.288, FWHM=.070) $ Transmission Coeff # 2 236 # of Transmission Pnts= 17 Transmission Coeff. = 0.0 237 # Filter - NO3306-6 (Cen = 3.3306, FWHM=.36) $ Transmission Coeff # 3 # of Transmission Pnts= 19 Transmission Coeff. = 0.100 Frequency (wavenumber) = Transmission Coeff. Frequency (wavenumber)= Transmission Coeff. Frequency (wavenumber) Transmission Coeff. Frequency (wavenumber) Transmission Coeff. # Filter - N0O3322-8 (Cen $ Transmission Coeff # 4 # of Transmission Pnts= Frequency (wavenumber)= Transmission Coeff. Frequency (wavenumber) Transmission Coeff. Frequency (wavenumber) Transmission Coeff. Frequency (wavenumber) Transmission Coeff. Frequency (wavenumber)= Transmission Coeff. Frequency (wavenumber)= Transmission Coeff. Frequency (wavenumber) = Transmission Coeff. Frequency (wavenumber)= Transmission Coeff. Frequency (wavenumber)= Transmission Coeff. Frequency (wavenumber)= Transmission Coeff. Frequency (wavenumber)= 238 3295 = 3.322, FWHM=.071) 18 Transmission Coeff. = 0.795 Frequency (wavenumber)= 3040 Transmission Coeff. 0.740 Frequency (wavenumber)= 3058 Transmission Coeff. = 0.100 Frequency (wavenumber)= 3071 Transmission Coeff. = 0.020 Frequency (wavenumber)= 3333 Transmission Coeff. = 0.0 # Filter - NO3411-6 (Cen = 3.411 , FWHM = .046 ) $ Transmission Coeff # 5 # of Transmission Pnts=12 Frequency (wavenumber)= 2000 Frequency (wavenumber)= 2942 Transmission Coeff. = 0.750 240 Frequency (wavenumber)= 2949 Frequency (wavenumber)= 2966 Transmission Coeff. = 0.100 Frequency (wavenumber)= 3333 Transmission Coeff. = 0.0 # Filter - NO3689-4 (Cen = 3.689 , FWHM = .046 ) # of Transmission Pnts= 18 Frequency (wavenumber)= 2000 Transmission Coeff. = 0.0 Frequency (wavenumber)= 2663 Transmission Coeff. = 0.0 Frequency (wavenumber)= 2667 Transmission Coeff. = 0.025 Frequency (wavenumber)= 2676 Frequency (wavenumber)= 2718 Transmission Coeff. = Transmission Coeff. = # Filter - NO3888-4 (Cen $ Transmission Coeff # 7 Transmission Coeff. = Frequency (wavenumber) Transmission Coeff. = 241 0.727 = 3.888, FWHM = .047) 15 Frequency (wavenumber)= 2576 Frequency (wavenumber)= 2587 Transmission Coeff. = 0.760 Frequency (wavenumber)= 2597 Transmission Coeff. = 0.100 Transmission Coeff. = 0.050 Frequency (wavenumber)= 3333 Transmission Coeff. = 0.0 # Filter - NO03990-4 (Cen = 3.990, FWHM = .190) $ Transmission Coeff # 8 # of Transmission Pnts= 17 Frequency (wavenumber)= 2000 Transmission Coeff. 0.0 Frequency (wavenumber)= 2381 It Transmission Coeff. 0.0 Transmission Coeff. = 0.100 Frequency (wavenumber)= 2503 Transmission Coeff. = Transmission Coeff. = # Filter - N04227-4 (Cen $ Transmission Coeff # 9 Transmission Coeff. = 243 0.800 = 4.227, FWHM = .055) 15 Frequency (wavenumber)= Frequency (wavenumber)= Transmission Coeff. Transmission Coeff. = # Filter - N04235-4 (Cen Transmission Coeff. = Frequency (wavenumber) Frequency (wavenumber)= 244 2339 = 4.235, FWHM = .181) 29 Transmission Coeff. Frequency (wavenumber) Transmission Coeff. = Frequency (wavenumber) Frequency (wavenumber)= Transmission Coeff. Transmission Coeff. = Frequency (wavenumber) Frequency (wavenumber)= 0.010 245 Transmission Coeff. = Transmission Coeff. = # Filter - N04461-8 (Cen Transmission Coeff. = 246 0.774 = 4.461, FWHM = .052) 21 Frequency (wavenumber)= Transmission Coeff. = Frequency (wavenumber) Transmission Coeff. = # Filter - N04577-4 (Cen $ Transmission Coeff #12 247 2225 = 4.577, FWHM = .260) # of Transmission Pnts= Transmission Coeff. = Frequency (wavenumber) iH Transmission Coeff. if Frequency (wavenumber) Transmission Coeff. = 17 # Filter - W04684-4 (Cen Transmission Coeff. = Frequency (wavenumber) Transmission Coeff. = = 4.684, FWHM = .600 ) 17 Frequency (wavenumber) Transmission Coeff. # Filter - N04693-8 (Cen $ Transmission Coeff #14 # of Transmission Pnts= Frequency (wavenumber) Frequency (wavenumber) 250 2439 = 4.693, FWHM = .167) 23 251 Transmission Coeff. = 0.775 Frequency (wavenumber)= 3333 Transmission Coeff. = 0.0 $ Transmission Coeff # Transmission Coeff. = 202 C.7 The Output Object Parameter File # Output Object (OUT) Parameter File # Simulation Name: Car # Created on Date: 9/3/97 # Output Parameters (General Information) (output.cxx) # Load General Output Parameters for Object labeled by id. Coordinate Output Type= Output File Index = Output File Parameters A + + + + FH Output File Param # 1 Output Coordinates = Output Incident Power = Output Photon Rate Pwr/Pht Rate in Log10 Output Detector Signal= Output Detector Noise Output Detector Total = Det. Output in Logi0 Load Specific Output file PArameters (based on index) Most answers are TRUE-1, FALSE-0O car_sim/data/car_move
The emissivity is the percentage of blackbody radiation that a gray body will radiate
at a particular spectral frequency and, according to the second principle of Kirchoff’s
Law, should be equal to the absorption coefficient. In some cases the emissivity may
be a function of temperature as well as frequency. The general “gray body” radiance
windows of the infrared regions (3-5 and 8-12 ym). Since most materials can be
approximated by a constant emissivity over these common wavelength ranges, IRIM-
AGE currently treats the emissivity as a constant with respect to frequency when
designed to simulate the atmosphere along the line of sight of a single detector. A
popular model of this kind is the LOWTRAN model. In an effort to correct for the
atmospheric effects in aircraft and satellite observations, the Air Force Geophysics
Lab (AFGL) developed LOWTRAN to accurately compute the transmission and ra-
diance characteristics of various atmospheric conditions using a band-model approach
with layer pressure as the only parameter. It is capable of modeling several common
gaseous pollutants as well as various aerosol backgrounds. LOWTRAN contains sev-
eral pre-defined atmospheric models as well as a method for defining a user-defined
atmosphere using multiple horizontal layers. Using LOWTRAN as a basis, the AFGL
derived MODTRAN [6] as a more accurate model for the infrared and visible regions.
MODTRAN improved the band model to include three parameters (pressure, tem-
perature and line width) all of which utilize a database of values derived from the
with the spectral bandwidth resolution of LOWTRAN being 20 cm! and MOD-
TRAN reducing it to 2 cm7!. The smaller bandwidth makes MODTRAN a superior
model in the infrared region of the spectrum. Since MODTRAN also retains all of
the major features and reliability of LOWTRAN, it is a good model to use as the
basis for simulating the atmosphere in the IRIMAGE environment.
tral transmission of a path in an atmosphere, but it also computes the spectral radi-
ance of the gases along the path and the background surface. Using the transmission
coefficient with the radiance calculations, MODTRAN can compute a total spectral
radiance value along a path. Although we discuss how we use MODTRAN to com-
pute the incident radiance upon the imaging system later, the following discusses how
MODTRAN computes the transmission coefficient, the background radiance, the at-
mospheric radiance and the final radiance along a specific path looking through the
band model in conjunction with its detailed parameter database. We will not discuss
the computation of the individual gas transmission coefficients because this is detailed
quite well by Anderson et al. [8]. One important note is that each individual gas
transmission coefficient for a specific layer depends upon the path length, 1, through
that layer. If we assume the transmission coefficient for each gas in an atmospheric
layer is computed, then the total atmospheric transmission coefficient for the n*” layer
transmission coefficient for the path. Again, the transmission coefficient along the
path is simply the product of all of the individual layer transmission coefficients
radiance of the surface, Lsu,¢(v). It assumes the background surface is a gray body,
opaque (Tsurf = 0) and is Lambertian in natue. Using (2.2.3) with temperature,
Tsurf, and albedo, p, for the surface, MODTRAN computes the spectral radiance of
Lsurg (Vv) = (1 — p)Lpp(Ts,v) where e = 1—p (2.2.6)
diation formula (2.2.1) in conjunction with the spectral transmittance of each layer.
As part of the calculations to determine each layer’s transmission coefficient, MOD-
TRAN computes the layer’s Curtis-Godson density weighted temperature, 6. This
temperature is part of the Curtis-Godson approximation which treats each atmo-
spheric layer as if it were homogenous in temperature and pressure [9, 10]. MOD-
TRAN extends this approximation to the radiance calculation by using @ in the
blackbody calculation. With all this in mind, MODTRAN uses the simple, spectral,
Latm(9,v) = Lgp(O,v)- (1 —7(v)) (2.2.7)
radiance contribution, L,, and then multiplying the radiance value by the product of
= Leen, v)(Tnalv) — Tn(v))
Using the various radiance equations and the total transmission coefficient, MOD-
considerably faster than a line-by-line model, MODTRAN is still computationally
intense. The calculations in conjunction with the necessary database access causes
MODTRAN to take up to several seconds to compute a single path. There is no
automatic procedure for defining multiple paths through the atmospheric model. Any
imaging simulation must supplement MODTRAN with external routines which define
individual paths and pass them to MODTRAN. In addition, MODTRAN can only
define an atmospheric model in terms of horizontal layers of gas. More complex
atmospheric models, i.e., blobs of polluted gas, require external routines to generate
specific profiles for each path. These routines must interpolate the path with the
atmosphere in order to generate an acceptable profile for MODTRAN. These problems
are solved by IRIMAGE.
accurate images, traces rays from a detector, through the optical system, into the
background scene to determine the exact sources of radiation as well as how each
source contributes to the overall intensity. In addition, ray tracing correctly models
the refractive and reflective properties of an optical system. These properties make it
ideal for determining the optical path through the atmosphere used by MODTRAN.
uses a probability function to generate a set of rays for each detector. In figure 2.1b,
we see that rays are fired off in different directions. Each detector might generate
just a few rays or several hundred rays. Since each ray is computed separately and
averaged with the others, the time involved is considerable. The accuracy of the model
depends on the probability function that defines the ray distribution and the number
of rays generated. Another method involves generating a single ray per detector and
selecting the most likely path it will take as seen in figure 2.1c. Although considerably
faster, this method is less accurate and depends greatly on determining the most likely
path.
lation. The user defines the path, i.e., ray, through the atmosphere with an opaque
surface at the end. Using the intercepted profile and the surface characteristics,
MODTRAN determines the radiance along that path. The drawback to MODTRAN
is its long computation time. Although a single ray calculation may take a fraction
of a second, when one considers that a 128x128 array has 16,384 detectors, it is clear
that a Monte Carlo method is not feasible. Therefore, IRIMAGE employs a single
ray per detector method with the possibility of sub dividing a detector into a set of
sub-areas. As shown in figure 2.1d, each sub-area also only represents a single ray,
Scene
Traced Ray
A Single Detector
Single Ray Method Single Ray Method Using Submesh
onto Background
onto Background
from Detector
Cc D
(b) Monte Carlo method. (c) Single ray per pixel method. (d) Single ray method
using sub-mesh.
a more accurate image.
to transform an incident ray on an optical system into an exiting ray, the single ray per
detector nature of the IRIMAGE calculation requires a less sophisticated approach.
The nodal points of an optical system are defined as the points on the optical axis
where a ray enters and exits with the same direction vector. If one assumes the
system obeys the paraxial approximation, then the most likely path for the ray can
be assumed to pass through the nodal points. Since a pin-hole camera follows this
same principle, we refer to this method as the Pin-Hole Approximation. The beauty
of this approximation is that the attenuation properties of the system are preserved
provided they are not spatially dependent (e.g., Fourier Optics). In addition, the
nodal points can be calculated using the transfer matrices of the optical elements
mentioned above. One simply solves for the point along the optical axis where the
incoming and outgoing rays have the same direction vector. There are numerous
optical design packages which will compute the cardinal points of an optical system
including the nodal points, so IRIMAGE does not compute them. It relies on the
the incident point on the background surface. Using this information, one can use
MODTRAN to compute the radiance value associated with the ray. However, this
value does not account for the orientation of the background surface, or the solid
angle between an element of the background surface and the surface of the aperture
of the optical system (see figure 2.2). These two quantities are necessary to compute
the radiation collected by the optical system and focused on the focal plane.
radiation onto surface (S2), the imaging system apeture. The ray from S$; to S5
—> Optical System
s (S,: Area=dS,)
( S,: Area=dS,)
then only the effective area of the radiating surface (dA) is needed. This effective
area is just the area projected on a plane perpendicular to the ray. Since the radiance
relies on the solid angle swept out by the receiving surface, then the solid angle must
also be determined. Using the effective area of the receiving surface (dA) and the
distance between the two surfaces (712), one can compute the solid angle. Using rio
requires that the distances from S; and S, to their respective projection planes is
Solid Angle (Sy to 52) -dQ = Aa _ So COS O5
be:
ray only determines the interception point on the radiating surface. The interception
point can provide the parameters for computing the radiance of the surface, but the
effective area seen by the detector is necessary to compute the power incident on each
detector. Figure 2.3a shows the single ray approach using a detector at a location
(x, y) on the focal plane. We project rays from the four corners of the detector through
the nodal points of the optical system onto the radiating surface. The interception
points define the “active” area, dS, on the radiating surface for the detector. This
area is assumed to have the same parameters as the point intercepted by the detection
ray, D. Using equation (2.2.11) and the angle between normal vector of the surface
Si, Sin, and D, we can compute the effective area of the radiating surface and
consequently the incident power on the optical system. This approach is independent
of the ray tracing method selected even though the detector area may be sub-divided
into smaller areas with a ray per sub-area.
collected power. However, there is a way to approximate the effective area of the
background surface seen by a detector using the detector’s own area. By doing so, we
can reduce the number of steps and computations significantly. The approximation
relies on the principle of similar triangles. In figure 2.3a, the rays projected from the
detector vertices onto the background sweep out a solid angle on either side of the
optical system. If the rays all pass through the nodal points, then by definition the
Single Ray
Element
Single Detector
Surface
Projected
Area
Background Asus CO8(%pd) AAs COS(Asu)
Surface Cr Oe nn ce: oe
(b) Using properties of rays passing through the nodal points, the solid angles swept
out on either side of the optical system should be equivalent. This fact allows us to
relate the visible surface area of the source and surface area of the detector using this
In order to determine the solid angle, we need the area swept out at a particular
radius value. On the source side, we can assume that the area swept out is equal to
the visible surface area determined in (2.2.11), ie., dAs = dAgury COS Osupp. AS can
be seen in figure 2.3b, the radius vector to the center of the solid angle area, rsurs, is
slightly longer than the radius vector to the center of the actual surface area which
is expressed aS Tsur¢ +dsurf¢. Similarly on the detector side, we use the visible surface
area of the detector, dAg = dAdget COS Osur¢, and the radius vector rage: + dae. Using
set eet fnew (2.2.16)
(rdet + dat) (surf + surf)
Tdet
If we take (2.2.17) and substitute it in (2.2.14), the resultant power collected by the
Paye(v) = L(v) AAs 208 Bact) Aapt C08 Bape) (2.2.18)
Since the sub-area ray tracing method is an extension of the single ray approach,
a more general form of (2.2.18) incorporates both methods. The sub-area method
divides each detector surface into a mesh of smaller rectangles. Each rectangle acts
area scheme. If a detector uses an N x M mesh, the general form of (2.2.18) becomes:
The single ray case is encompassed by this general equation where N = M = 1.
Once the spectral power is collected by the aperture, it can be adjusted by the
optical system. As stated in section 2.2.3, the optical system is capable of attenu-
ating the incident spectral power through the use of a spectral transmission curve,
Topi(v). The transmission curve provides the transmission coefficient for any spectral
frequency. Thus, the attenuated spectral power at a frequency v is the product of the
Paaj(v) = Topt(v) Pape (v) (2.2.20)
together to form a piece-wise continuous curve. The simulation derives the value at
any point on the transmission curve by linearly interpolating between the two known
data points that straddle the desired frequency.
sorbed by the optical system or incident on the detector, then (2.2.20) in conjunction
with (2.2.18) and (2.2.19) describe the spectral power incident on a given detector.
Once we know the incident power, it is possible to compute the detector signal and
the detector response function, i.e., the Responsivity. The response of any detector
may or may not be a function of the frequency. For most photon detectors, the
frequency. However, since there is some dependency, equation (2.2.21) demonstrates
how the output signal of a photon detector is simply the convolution of the spectral
power function, P(v), and the responsivity, R(v), of the detector. Thermal detectors
rarely depend on the spectral nature of the incident radiation. Instead they simply
respond to the change in temperature of the active material which is a result of
the total incident power. Therefore, equation (2.2.22) is just the product of the
the spectral power values in terms of a set of discreet spectral lines. The user defines
the resolution of the computed power spectrum by setting an integer number of
wavenumbers between each computed spectral line. The user must also define the
minimum wave number, Mn, and the number of lines to compute, N. Using these
factors, MODTRAN computes the average power per cm™! for each spectral line.
The discreet nature of the returned computed power reduces equations (2.2.21) and
Photon Detector: Vsig = S> R(v)P(v)Av (2.2.23)
ant NAv
Thermal Detector: Vsig= R S> P(v)Av (2.2.24)
simulation or experimentally. In either case, the output should be a set of data points
which may be frequency dependent. If this data could be fit to a specific function, then
one would simply need to plug that function into (2.2.23) or (2.2.24) and sum over
the various frequencies. However, since various detectors may have their own unique
fitting function, it would be impractical to recode the simulation for each individul
each point is joined to the adjacent points by lines. Figure 2.4a demonstrates how a
response curve can be made up of points connected by lines.
any two points on the curve. Numerically, the responsivity is determined by either
finding two known data points whose frequencies either straddle the given frequency
or by finding a data point that shares the same frequency. In figure 2.4b, a given
frequency straddles two frequencies. Using (2.2.25), the desired responsivity is found
detectors much easier. The algorithm to compute R(vp) consists of a search routine
to find the two straddle points followed by using the above equation or finding an
equivalence point. This method allows IRIMAGE to define a generic response algo-
rithm instead of a series of detector specific, response algorithms. In addition, this
algorithm is based on actual computed or measured data instead of relying on the
user to write their own detector response routines. However, as we will discuss in
the next section, the object-oriented nature of IRIMAGE still allows a user to create
cision does not always mean accuracy. In the case of detecting IR radiation, all of
the theory presented up to this point ignores background fluctuations of the incident
radiation. In order to account for this random behavior, the simulation must include
a way of modeling these fluctuations.
Data Points
oe. , | CS.
> 3 ' \ Responsivity
BS 1 \ rve
ap \
os i
pa t----pp----
me
_\*— Spectral Band \
» ~~
r 1 1 l 1 1 1 a ed |
5 Riv) eee (Vo - Va) + R(va)
Pa
id » —
ab . _--&
a § 7 RR)
S 2
ae Riv.)
I | I rT I
B Spectral Units
straight lines. (b) Interpolating the responsivity of a detector for a spectral band with
frequency, ¥, and bandwidth, Av, that lies between two points on the responsivity
curve.
from the background. These variations may come from the background scene, the
atmosphere and even the optical system. We define that the mean number of photons
incident upon a detector with surface area Ap and integration time tin, is N. As-
suming that the fluctuations follow a random-walk statistical pattern, then equation
(2.2.26) describes what the mean square fluctuations of incident photons, (AN)?,
= S- (AN(v))?(Av/)? (2.2.27)
been eliminated.
generator to generate realistic random noise voltages. In order to generate the noise
probability function, the standard deviation of the noise voltage is necessary. Using
the results of (2.2.27), we compute the standard deviation of the noise voltage.
= R(v)[P(v) — P@)|Av
a (2.2.28)
= R(v)hv|N(v) — N(v)|Av
= S° R(v)hv[N(v) — N(v)]Av
Squaring and averaging equation (2.2.29) results in the following equation:
(AV noise)? = (92 R(v)hy[N(v) — N(v)|Av)?
| (2.2.30)
two cases.
v a (2.2.31)
= S$°(R(v)hv)?N(v) Av
mean voltage. Using any normalized probability function and the noise voltage stan-
dard deviation, one can generate a random distribution of noise voltages that is
weighted by the probability function through the use of the probability density func-
tion. In order to understand this technique, we must revisit the Probability Density
F(a) =[ P(x) 5x (2.2.32)
than or equal to a. For a normalized probability function, P(x), the value of F(a)
must be between 0 and 1. If we assume that a and x are voltages and P(z) is the noise
probability function defined by a mean noise voltage, V;,, and the standard deviation,
On = (AV)?, then one can determine the probability density for any voltage.
want to achieve, then we can solve for the value of a that will generate this probability
density. Most computer algorithms generate random numbers with a flat distribution
between 0 and 1. However, if one defined the value of the probability density using
a number generated by one of these alogithms, then the resultant alpha would be a
random value that is weighted by P(x). We can demonstrate this graphically. First,
Function Bins
(A)
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
(B)
tion using a Reimann sum. (a) Normal representation of P(x) with equally spaced
divisions. (b) Number line with regions representing the area of each equally spaced
division.
a normalized Gaussian probability function where the area of each rectangle defines
AF‘(2z;). The centers of each rectangle are evenly spaced by the value Az and the
center of rectangle n is defined by tp = 2min + nAz. In figure 2.5b, each AF (a;)
is stacked sequentially along a number line from 0 to 1, effectively summing over all
of the AF(z;) values. By doing so, each AF'(x;) defines a small but unique region
of the number line which corresponds to the value z;. Therefore, given a specific
F(a) which defines a point on the number line, we would simply determine which
DeltaF(a;) region contains this point and set a equal to the corresponding 2;,.
is proportional to the probability of having the value z;. In the figure, a value of zx
an x value near one of the edges. In general, for regions of high probability, a large
range of F(x) will correspond to a small range of x values, while in regions of low
probability, the exact opposite is the case. Therefore, a random value F(a) will have
an a which is weighted by the probability function, P(z).
Using a numerical algorithm based on the number line method above and the
results of equation (2.2.31), IRIMAGE can generate a realistic random noise voltage
1. Compute standard deviation of the noise voltage, oy, using (2.2.31).
2. Define or compute the following parameters for the probability density sum:
e The voltage range: —Ao, <= x <= +Ao, where A > 0 and areal number.
e The number of AF'(x) regions to sum over: N.
3. Plug oy and y into the Gaussian probability function, Peauss(x).
4. Compute each AF(x;) and store each AF (z) in an array.
5. Sum over all of the AF(z;) and normalize them so that sum equals 1.
6. Use a random number generator to get a number, Fo, between 0 and 1.
vantages to setting the mean of the probability function to zero. With a zero mean,
Peauss(Z) is symetric around zero making the range symetric around zero as well
which simplifies the calculation. In addition, the mean only effects the location of
the central point of Peauss(z) and not the shape. Thus, the only effect would be that
after the noise calculation. Therefore, a zero mean does not effect the total voltage
and allows us to separate signal and noise voltages quickly.
determines how much of the probability function to sum over while N determines
the resolution of the approximation. If the range is too small, then the result will be
skewed toward a more even distribution and not a Gaussian. However, if the range is
too large and N is too small, then the resolution of the calculation may be too rough
which could also alter the distribution and could produce unpredictable results. Even
for a reasonable range, if N is too small, then it limits the most probable noise voltages
to only a few noise voltages since only a few are in the high probability portion of
the distribution while the rest are in the low probability regions. On the other hand,
the larger N becomes, the slower the calculation takes. Therefore, a value of A = 2 is
sufficiently large enough to encompass most of the Gaussian and a value of N = 1000
simulation. Due to the object-oriented approach used by IRIMAGE, it is beneficial to
define certain computational constructs and programming objects which will simplify
the description of the different simulation objects. Since any imaging simulation is
heavily dependent upon the positions of various objects, we briefly introduce the
World Coordinate System (WCS) used by IRIMAGE. Furthermore, the Versatile
Grid Object (VGO) is an integral part of the Scene, Atmosphere and FPA objects
which makes it necessary to discuss it as well as the methods the VGO uses to
manipulate both two and three-dimensional data. Finally, we introduce the Ray
program object which stores the information generated by the backwards ray tracing
method. Although many of these concepts are introduced in appendix B, they are
System
of Optical System
(Origin of WCS)
a right-handed coordinate system whose origin lies along the optical axis exiting the
imaging system and sits on the front surface of the optical system. The +Z axis
points away from the imaging system along optical axis. The +X and +Y axis are
oriented to be horizontal and vertical respectively and follow the right-handed nature
of the WCS.
ment of different simulation elements, both geometric models and simulation objects,
in three-dimensional space. Some of these elements may even maintain an internal
coordinate system. However, IRIMAGE defines a single primary coordinate system
called the World Coordinate System (WCS). The WCS is a right-handed coordinate
system with the positive Z axis pointing along the optical axis and the Y axis pointing
up. Its origin is at the point where the optical axis and the external surface of the
optical system intersect. Figure 2.6 shows the location and orientation of the WCS
with respect to the imaging system.
defined in a coordinate system that can transform points between itself and the WCS.
Therefore, IRIMAGE restricts each element coordinate system (ECS) to be right-
handed, orthogonal and share the same unit of length as the WCS. In addition, any
element that uses an ECS must maintain a set of transformation matrices. These
matrices allow a point defined in the ECS to be transformed between the ECS and
the WCS. Since the transformation matrices may have to rotate and translate points
from one coordinate system to another, the order of the rotations and translations
are very important. We define the rotation and translation matrix for a point going
from an ECS to the WCS using equation (2.3.1). Since transforming the point back
should return it to the same position in the ECS, the other transformation matrix is
Mwes—rcs = T™*R,*RS*R;z* (2.3.2)
transformed into a different ECS using the WCS as a bridge. Equations (2.3.3)
and (2.3.4) demonstrates how a point, P, is transformed from EC'S; to the WC'S and
then could be transformed from the WC'S to another element’s coordinate system,
storing spatially oriented integer or floating point information. While the VGO is
inherently three dimensional, its structure is capable of being one, two or three-
dimensional. It does maintain its own ECS which is called the Grid Coordinate
System (GCS). As shown in figure 2.7, the origin of the GCS lies at the physical
center of the grid structure. Using the origin of the GCS and the physical spacing
between adjacent grid points, the VGO can determine the spatial coordinates of any
grid point in terms of the GCS.
GCS and the WCS. In order to utilize these equations, the VGO requires that the
GCS origin and the orientation of the GCS about this origin be defined with respect
to the WCS. Using these values to generate the transformation matrices, the VGO
can then relate any coordinate in the WCS to one in the GCS.
the size and data type of the array isn’t defined until the VGO is first initialized.
Although flexible, the VGO requires every grid point array to be the same size and
data type. This requirement avoids a lot of bookkeeping problems. In addition to the
data array, each grid point stores a single “source of data” index. Using this index
allows the VGO to not only store data from various sources onto the grid, but it allows
the VGO to associate a unique index value with the data from a particular source.
Using this index, the VGO, or an external routine, can limit data manipulation,
retrieval or other operations to data from a particular source.
simulation object represents the background image surface by mapping temperature
and/or emissivity image data onto a 2D form of the VGO. In this case, each grid point
simply acts like a pixel of an image except instead of storing a color value, it stores
a temperature and emissivity value. Another two-dimensional form of the VGO is
used by the FPA simulation object. Since a focal plane array already consists of a
grid of detectors, the VGO acts as the natural data structure to store information
about each detector in the array. Finally, the Atmosphere simulation object puts a
three-dimensional form of the VGO to extensive use. The VGO provides a way of
storing three-dimensional volumetric data in a form that the ray tracing technique can
translate into an atmospheric profile for use with MODTRAN. These three simulation
objects utilize the VGO as a major part of their operation which makes it an excellent
example of the benefits of code reuse.
the VGO does not perform data interpolation. It only loads data directly into the
grid point arrays using the spatial coordinates for the data. Actual interpolation of
any image or volumetric data onto the grid is left to external routines instead. In
the next few sub sections, we discuss how specific two and three-dimensional data is
2.3.3 Interpolating 2D Image Data onto the VGO
model by placing temperature and emissivity images onto a 2D VGO. In general, the
method for interpolating image data onto a 2D grid involves mapping the coordinates
of each data point onto the grid plane and then using a standard 2D interpolation
algorithm to determine the value of each grid point. In IRIMAGE, a program object
called a Virtual Image Object (VIO) maps the image coordinates onto the image plane
and a near point or bilinear interpolation technique determines the value of each grid
point. Using this method, any number of images can be loaded onto the VGO in
a mosaic form. However, one can also combine objects using a “matte” process in
conjunction with another program object called the Layout Object. With this added
technique, the Scene object can create fairly sophisticated images to represent the
background scene using just a few simple images.
each image used in the simulation is loaded into a database and is placed on the
imaging grid using a set of Virtual Image Objects. Each VIO is assigned a particular
image and a set of image processing operations which change the placement, scale,
or rotation of the image before it is placed on the imaging grid. For simplicity, all
of these operations use the Grid Coordinate System of the Scene object instead of
the World Coordinate System of the simulation. Figure 2.8 shows an example of an
image being scaled and rotated by a VIO. Since a single image may be associated
with several VIO’s, each VIO processes its image and then returns it to its original
state. This eliminates the need for storing the same image multiple times.
makes use of a key frame database to store the parameters used to process its as-
sociated image. Each key frame defines the active state and the image processing
parameters of the VIO for a specific frame of the animation. The active state des-
Image from VIO
Current State Routines Scaled & Rotated
of Background Image on Background
Imaging VGO Imaging VGO
VGO. Otherwise, it skips the VIO entirely. Since the active state is just a flag, the
active state only changes if another key frame changes it. For the case of a frame
that lies between two key frames and the VIO is active, it will compute the image
processing parameters by linearly interpolating between the two key frames. The
VIO only adjusts its associated image while it interpolates onto the Scene VGO and
its original state once this step is complete. By using a series of VIO structures
with their associated key frame databases, the Scene simulation object can simulate
a dynamically changing and interesting background surface.
Mattes and The Layout Object
are composited together. Since most images are inherently rectangular, each image
uses a “soft” matte to combine two images using a simple weighted average. Such
a method is useful for anti-aliasing the edges of the combination as well as making
any part an image appear to be semi-transparent. The compositing method used by
IRIMAGE treats a matte image much like any other image by associating it with
a VIO. However, a matte VIO interpolates the matte image data onto the Scene’s
imaging VGO using a separate matte variable, or “channel.” This separate channel
makes it possible to combine an image already loaded on the imaging VGO with a
new image.
the imaging VGO. Figure 2.9 shows a car composited onto a background image. If
the matte image were not loaded prior to the car image, then the entire car image
would have been loaded, blocking out the background. In order to preserve the
image loading order, the Scene simulation object employs a Layout program object
which stores the order of VIO’s loaded for a particular frame. Since the VIO’s can
be animated, the Layout Object must also be capable of being animated. The key
frame database used by the VIO is also used by the Layout object. However, instead
of storing processing parameters, it simply stores the VIO loading order. Like the
active state in a VIO, the loading order is not interpolated but is set until the next
key frame changes it. Through an effective use of matte’s and layout order, the
Scene simulation object can create a sophisticated background image from just a few
interesting temperature/emissivity images.
VGO and the GCS of the interpolated image’s VGO. Although it would seem logical
to project the image VGO onto the Scene VGO, it actually works better in reverse.
By projecting the Scene VGO onto the image VGO, each grid point in the Scene
VGO that lies within a rectangle formed by four points of the image VGO should
be interpolated. All other points are considered to be outside of the image and
Background Imaging VGO
On Background Imaging VGO
Background Imaging VGO
Background Imaging VGO
the background image. (d) Final composite image of car on background.
surrounding each Scene grid point makes it easy to compute either the nearest point
value or bilinear interpolated value.
grid points. Using the Near Point approximation, this interpolated grid point assumes
the value of the nearest image grid point (P = .7). This is a quick and dirty method
of interpolating points. However, it relies on the fact that the values do not change
rapidly between grid points. The Bilinear method [13] utilizes a weighted average
between the four image grid points that surround each projected Scene grid point.
One can see in figure 2.10d the method projects a line parallel to the X axis and one
parallel to the Y axis which pass through the point to be interpolated. By computing
the area of each small rectangle as a percentage of the original rectangle’s area, the
method multiplies the grid point value by its associated rectangle area and sums the
four values up. The result is the bilinear interpolation of the four grid values. This
method is more accurate because the actual location of P effects its value, but it
using temperature or emissivity image data. The advantage of a VGO solution for
modeling the 3D atmospheric data lies in the benefits of a Ray-Grid interception
technique over a Ray-Geometric Model (GM) technique. For each frame of the Ray-
GM approach, each ray checks to see if it intercepts every GM in a the scene. Since
only a few GMs in a scene will be along a specific ray’s path, a lot of CPU cycles
may be wasted checking GMs that will not be hit. However, if one subdivides space
into a regular, orthogonal, set of grid boxes and interpolates each GM onto the grid,
then most of this wasted time can be averted. In this Ray-Grid case, each ray simply
checks each grid box it passes through to determine if part of the GM lies within the
grid box. When part of a GM does occupy a grid box, then the method checks the
Ray-GM intersection. If the ray does intersect the GM, the search is over. Otherwise,
A Coordinate System B Coordinate System
V=0.2 V=0.7
the image grid on the Scene VGO. (b) Projecting the Scene VGO onto the image grid.
(c) The Nearest Point interpolation method. (d) The Bilinear interpolation method.
the search continues. Various methods like this have been introduced in the past that
clearly show a performance improvement by using a grid approach to ray tracing in
computer graphics [14, 15].
GM intersection faster and still require intersecting the ray with the GM. In the case
of IRIMAGE, the GMs we deal with are transmissive which make it necessary to
continue checking for Ray-GM interactions even after one is found. The checking
process goes on until the ray strikes the Scene object’s imaging plane. Furthermore,
since each GM stores volumetric data, any ray intersecting a GM needs to store the
volumetric data in addition to the length of the ray that passes through the GM.
The ray must also store data for the regions where a GM does not exist by storing a
default set of values and the corresponding path lengths. However, by interpolating
these GM’s onto a three-dimensional VGO, we can significantly improve this process
of checking and storage of data. The Atmosphere simulation object assigns each GM
a unique index before it is interpolated on the VGO. As each GM is interpolated,
any grid point inside the bounding surface of the GM is assigned the GM’s index.
As each ray intersects the VGO, it simply keeps track of which grid boxes it passes
through and the path length through each box. Once it completely intersects the
grid, it can then cycle through the stack of grid boxes and retrieve the indexes for
each grid box. Using these indexes and corresponding path lengths, the ray can build
an atmospheric profile from the associated GM’s atmospheric conditions as well as
any default conditions. This modified Ray-Grid intersection method eliminates the
need for keeping track of whether the one GM is inside another as well as where a
GM ends and the atmosphere begins.
mensonal VGO. In IRIMAGE, each GM is defined as a GridObj program object. The
GridObj uses a bounded surface approach based on the Universal Primitive Object
(UPO) (see appendix B). The UPO is a hybrid, edge-based, surface representation
well suited for defining non-intersecting, bounded surfaces. Its hybrid structure is
in its own coordinate system and it only stores a single unique object index. The
GridObj adds the ability to animate the placement, orientation and scale of a UPO’s
bounded surface in the WCS. It uses the same Key Frame database object used by
the VIO’s in the Scene simulation object. Since the volumetric data associated with a
GM may change over time, each key frame also stores the volumetric data associated
with that frame. All of these attributes allow a GM to be completely defined using
the GridObj as well as prepare it to be interpolated onto a VGO.
and dice” algorithm defined by Springfield to interpolate its UPO-based bounded
surface onto the VGO. This algorithm defines each grid point of the grid structure
to be surrounded by a bounding box. If the algorithm finds more than 50% of the
grid box’s volume to be inside the GM, then GM’s volumetric data is associated
with the grid point through the grid point’s unique index value. Using the GridObj
interpolation algorithm, any GM based on the GridObj can build itself, transform
its bounded surface to the WCS, be associated with a specific set of volumetric data
and then use the bounded surface to interpolate its associated volumetric data onto
three-dimensional Atmosphere VGO and the two-dimensional Scene VGO. Each Ray
stores the local origin, a local direction vector and its current endpoint in the WCS.
The local origin establishes starting point of the Ray. However, the location of the
starting point can be adjusted by another object, such as the Optics simulation object,
so that the value stored by the local origin always defines the current starting point
of the Ray. In IRIMAGE, this usually begins as the center of a detector on the FPA,
but is replaced by the exit point of the optics after the Ray is processed by the Optics
object. The direction vector defines the current direction that the Ray points. This
vector may also be adjusted by the Optics object. It is used to compute points of
interception between the Ray and any surface. The endpoint stores the coordinates of
the point where the ray intercepts the background surface, i.e., Scene object’s imaging
VGO. Using the endpoint, the Scene object can retreive the surface temperature and
emissivity from the imaging VGO.
indexes and path lengths. The Atmosphere VGO can be defined as a set. of grid
boxes, or voxels, where each point in the grid lies at the center of one of these voxels.
As the Ray goes from the exit point of the optics to its endpoint on the background
surface, it will pass through the Atmosphere VGO and storing the grid point index
to every voxel through which it passes. The Ray-VGO interception algorithm forces
these indexes to be stored in the proper order to build an atmospheric profile. As the
Ray passes through each voxel, it computes the length of path that it travels in that
particular voxel and stores it as well. Using the path lengths in conjunction with the
grid point indexes, the Atmosphere object can then use the Ray grid point database
describe each of the major objects that make up IRIMAGE and how they contribute
to the overall simulation. In section 2.5, we will put these objects together to describe
scene. As mentioned in section 2.3, three-dimensional IR, scene generators have been
in existence for over 10 years. The object-oriented nature of IRIMAGE and specif-
ically of the Scene simulation object will allow us to incorporate those models into
a basic scene generator, makes a two-dimensional background model more practical
in this initial stage of IRIMAGE.
various temperature and emissivity “images” onto a two-dimensional form of the
VGO. Since the imaging VGO is what defines the background scene, it must be
placed properly so that it can be “seen” by the imaging system. Therefore, the user
must specify its size, location and orientation in WCS as well as the number of grid
points along the two orthogonal axis of the grid. The VGO also needs the default
temperature and emissivity values so that no grid point has these values undefined.
Once these parameters are set, the imaging VGO is ready to be loaded with data.
separate two dimensional VGO. Each image may be generated from the output of a
real IR imager, from a 3D synthetic image generation, or using a standard graphics
package. IRIMAGE supports 8 bit and 24 bit portable bitmap formats [16] as well as
ASCII files with a single column of data. Because the bitmaps are in binary, the user
must specify the gain and offset values required to convert each 8 bit number into a
temperature, emissivity or matte value. In addition, IRIMAGE treats the red, green
and blue channels of the 24 bit bitmap as three inidividual 8 bit images. By doing
so, it is possible to store the temperature, emissivity and matte images in a single
bitmap. Since ASCII is capable of storing floating point information, the data must
already be the actual temperature, emissivity or matte values. Once these images
have been loaded, the Scene object can associate them with the VIO’s.
by the database of VIOs and the Layout program object. For each frame in the
simulation, the Scene object accesses the Layout object to generate the ordered list
of VIOs to load onto the imaging VGO. Each VIO determines the image processing
parameters using the current frame. Once the parameters are defined, the image data
linked to the VIO is manipulated, loaded onto the imaging VGO and then returned to
normal. Matte images are loaded into the imaging VGO’s matte channel. Tempera-
image data using the matte channel. To prevent the matte from effecting more than
one set of image data, the Scene object resets the matte channel after each compos-
ite. Once all of the images are loaded on the imaging VGO for the current frame, the
Scene object is ready for the ray trace.
the background temperature and emissivity for the corresponding ray. Since IRIM-
AGE does not allow any VGO to change its position, size or orientation between
frames, the coordinates of the interception point between the ray and imaging VGO
remain the same throughout the course of the simulation. By storing the intercep-
tion coordinate in the Ray object instead of the temperature and emissivity values,
each Ray only needs to compute this interception at the beginning of the simulation.
Consequently, the Ray may retrieve the background temperature and emissivity for
any frame by submitting the interception coordinate to the Scene object which will
quickly retrieve the values from the imaging VGO. Although the calculations in-
volved are simple, the shear number of rays generated per frame make any reduction
and uses the atmospheric modeling code, MODTRAN, to compute the spectral ra-
diance of the background surface and intervening atmosphere. While MODTRAN
only provides a one-dimensional layer solution, the Atmosphere object uses the VGO
combined with a Geometric Model Database (GMD) object to define the atmospheric
conditions three-dimensionally. The one-dimensional MODTRAN profile is then gen-
erated using the simple Ray-Grid interception technique. When combined with the
surface data from the Scene object, the radiance incident on the optical system can
be calculated by MODTRAN.
geometric model index (GMI) and two integer indexes which relate the point to a gas
model and an aerosol model. The GMI links an interpolated volumetric data set with
each grid point that lies within its bounding surface. The gas model index points to
an element of the Gas Model Database which contains an overall temperature and
pressure as well as a set of specific concentrations for each of the gaseous constituents
modeled by MODTRAN. Similarly, the aerosol model index points to an element of
the Aerosol Model Database which stores the aerosol modeling parameters used by
MODTRAN to incorporate aerosols into the atmospheric model. Since both of these
databases are linked to MODTRAN, they are stored in the MODTRAN program
object, a program object contained inside the Atmosphere simulation object. The
VGO does not need to cover the entire region between the imaging system and the
Scene object. Instead, it is only required in the region which needs to model the
atmosphere using volumetric data. The rest of the space is defined using the default
gas and aerosol model indexes.
databases of geometric models which are based on the GridObj program object. Each
database in the GMD stores all of the geometric models associated with one of four
standard bounding surface types: a box, a cylinder, a cone and an ellipse. The user
defines each geometric model and the gas and aerosol indexes associated with it.
Once the individual databases are loaded, the GMD can load the geometric models
onto the Atmosphere VGO. Like the Scene object loaded the VIO’s, the order of
how the GMs are placed on the Atmosphere VGO is very important. Therefore,
the GMD uses the same Layout program object used by the Scene object to define
the layout order of the geometric models. The Layout Object stores each geomtric
model’s primitive surface type as well as its index in the primitive surface database.
This avoids confusing two different primitive types which have the same index in their
respective databases. In addition, more complicated geometric models that represents
a specific set of atmospheric conditions can be generated through the effective use of
its own coordinate system and then transformed to the WCS using its own key frame
database. Once the GM is defined in the WCS, it is transformed into the Atmosphere
VGO’s GCS and interpolated on the VGO using the GridObj interpolation routine.
uses the grid point indexes in conjunction with the gas and aerosol databases to
compute the atmospheric profile. This profile is passed onto the MODTRAN object
to compute the spectral radiance. Since the Atmosphere VGO does not need to
occupy the entire space between the optics and the background surface, the Ray-Grid
interpolation requires three major steps. First, it determines the distance the Ray
travels in the atmosphere between the imaging system and the Grid. If this distance
is greater than zero, it stores the distance and the grid point index (—1, —1, —1) which
tells the ray to use the default atmospheric model indexes. Second, the Ray intersects
each grid box along its path storing the grid point index of the grid box (not the
atmospheric model indexes) as well as the distance the ray travels through the box.
As we show two-dimensionally in figure 2.11, the intersection technique represents
each grid box using a series of parallel planes that lie along the three major axis of
the GCS. Each plane is separated by the grid spacing along each axis. Once the entry
point on the first plane, i.e., entry point of the grid, is found, the method checks
the three orthogonal planes nearest this point along the direction vector of the ray.
It selects the next point by finding which of the next set of planes intersect the ray
closest to the first point. The path length through the grid box is simply the distance
between these two points and the grid point index is determined by the VGO using
the mid point between the two intersection points. In the third step, if the ray has
not struck the Scene object’s imaging plane before it exits the Atmosphere VGO, the
ray again stores the (—1, —1, —1) grid point index and computes the distance between
the Atmosphere VGO exit point and the interception point with the Scene’s imaging
VGO. Figure 2.12 demonstrates these three steps.
Represented by Gray Grid
Ys
Planes
Bounding
Planes
going through the stack of grid point indexes and accessing the Atmospheric VGO
to get the atmospheric model indexes. When consecutive grid boxes have the same
gas and aerosol indexes, they are combined into a single layer and their path lengths
are added together. Using this method, the atmospheric profile along any ray is built
and passed on to MODTRAN.
dexes in the Ray object. Instead, these are retreived from the Atmosphere VGO when
it builds the atmospheric profile. If a geometric model changed in anyway between
two frames, it is reinterpolated on the VGO. However, this only changes atmospheric
model index values that are stored in the Atmosphere VGO. Consequently, the atmo-
spheric profile would change, but the grid point indexes and path lengths that build
the profile would remain unchanged. Therefore, as long as the Atmosphere VGO
remains stationary in space, IRIMAGE only needs to compute the Ray-Grid inter-
vVGO
VGO
the Backwards Ray Trace.
and intercepts Grid Boundry.
Imaging System and Grid Boundry.
Step 2: Intercept Ray with each Step 3: Ray exits Grid and
Grid Box along the Ray path Intercepts Background Scene Grid.
Compute length (1,) of Ray in each Compute distance from Atmosphere
c Grid Box and store Index. D Grid to Background Scene Grid.
frames.
up the atmospheric profile for each ray and then passes the profile to the MODTRAN
model which computes and returns the spectral radiance for the ray. This object loads
the general parameters that define the MODTRAN model as well as the individual
gas and aerosol models used by the geometric models stored in the GMD. It also
processes the atmospheric profile generated by the Atmosphere object and returns a
MODTRAN compliant profile. The MODTRAN object acts as the connection point
between the C++ code of the Atmosphere object and the FORTRAN code that
defines MODTRAN. Using this link, the Atmosphere object passes a complete profile
to the MODTRAN object which passes it through a set of linking routines to the
MODTRAN code. After performing the computation, the MODTRAN code returns
the spectral radiance back through the linking routines to a spectral radiance array
which is passed onto the imaging simulation objects in IRIMAGE.
other atmospheric modeling codes. For this reason, all of the data structures and
routines related directly to MODTRAN are stored in the MODTRAN program ob-
ject. This separates MODTRAN from the rest of the Atmospheric simulation object.
However, many of the routines in the Atmospheric object are still based heavily on
the MODTRAN methodology. Therefore, new atmospheric codes must conform to
the MODTRAN methodology or the Atmospheric object will need to be adjusted to
conform to the new methodology without changing how data is passed into and out
Once all of the major simulation objects load their parameters and the Scene and
Atmosphere objects are initialized, the FPA object lays out a two-dimensional VGO
stack. These rays are passed on to the Optics, Scene and Atmosphere simulation
objects which return the spectral radiance of the background scene and intervening
atmosphere. Once the radiance incident on the detector stack is returned, the FPA
passes it to the Detector simulation object which computes the output value for a
given detector. The output value from the Detector object is finally passed to the
Output simulation object which places each detector’s output into a data file. All of
this interaction with the other simulation objects makes the FPA the only simulation
object which is heavily dependent on the structure of these other objects.
ion onto the focal plane. Like the other two simulation objects that use the VGO
object, the user defines the location and orientation of the FPA grid with respect
to the WCS. In this case, the origin and orientation are usually centered on the op-
tical axis with the two-dimensional grid perpendicular to the axis. The number of
grid points is specified by the number of detectors along the X,Y directions and the
spacing is defined by the detector pitch in both directions. The UnitCell program
object “tiles” a specific pattern of detection elements onto the FPA’s VGO. The user
defines a detection element by linking it to a specific set of detectors stored in the
Detector object’s database. Each detection element represents a detector stack by
storing the type and index value for each detector in the stack. In addition, it stores
the spectral range and bandwidth of the stack. Once a detection element is defined,
it is assigned a unique index and stored in the UnitCell object. The pattern defined
by the UnitCell object can either be a set of completely different detection elements
or a few detection elements in a specific repeated order. Using the pattern of indexes
as a guide, the UnitCell object associates each grid point on the VGO with a specific
detection element. In figure 2.13, we show an example of a four detector unit cell as
it is tiled across the surface of the focal plane grid to form an 8x8 FPA. When the
FPA object needs the detectors or information for a particular grid point, it uses the
detection element’s index to access the information from the UnitCell object. There-
(Made up of 16 Unit Cells) (UnitCell)
Detection Elements
Ray objects associated with each grid point of the VGO. The user defines the detection
area as a specific percentage of the product of the rectangular area that surrounds
each grid point. In addition, the user may select either the single ray or sub-mesh
approach to assign Ray objects to each detection area. For the single ray method, the
FPA object places each Ray object’s origin at the center of the detection area which
is just the associated grid point location in the WCS. The mesh method divides the
detection area into an n x m array of rectangles where n and m are the user defined
number of rectangles along the X and Y directions. The origin of each Ray object is
placed at the center point of each smaller rectangle. Once the Ray objects are created,
the FPA object uses the back nodal point of the optics with each Ray object’s origin
to compute the initial direction vector of the ray. The two approaches converge after
the Ray objects are defined.
origin is replaced by the exit point from the external surface of the optics while the
direction vector remains the same. Each Ray is returned by the Optics to the FPA
where it is handed off to the Scene object to determine the interception point with
background VGO and to the Atmosphere object to build its Ray-VGO interception
database. After all of the rays are processed, IRIMAGE proceeds to compute a
single frame or an entire animation. During each frame, the FPA object passes the
rays to the Atmosphere object to generate the atmospheric profile along the ray
and uses the Scene interception point to set the background surface temperature
and emissivity for the profile. Before passing the final atmospheric profile back to the
Atmosphere object to be processed by MODTRAN, the FPA accesses the Ray object’s
associated detection element in the UnitCell program object to get the spectral range
and bandwidth for the profile. This final set of parameters define the active spectral
region of the detectors in the stack. Once the profile is completely generated, it
is passed to MODTRAN, via the Atmosphere object, which returns the spectral
radiance of that profile. Finally, the FPA object passes the spectral radiance of the
Ray object back to the Optics object where the transmission curves are used to adjust
the spectral radiance. The attenuated spectral radiance is returned to the FPA to be
converted to the spectral power incident upon the detection element.
spectral power. The single ray method calculates the incident spectral power using
the spectral radiance of the ray and the area of the entire detection element. The mesh
method also uses the spectral radiance of each ray, but the area is reduced to the area
of the rectangle. The mesh method then sums the spectral power incident on each of
the rectangles to generate the total incident spectral power on the detection element.
The actual computation is the same in both cases, but the mesh method accounts for
the multiple rays involved due to the subdivision of the detection element.
output signal for the detectors of that element is calculated and stored in the output
to the Detector simulation object. It uses the type and index values to access the
correct routines and parameters for each detector. Once the right detector is found,
Detector object converts the incident spectral power into an output signal as well as
computing the background fluctuation noise. The final output values are returned to
the FPA object which submits it to the Output object for storing in an ASCII data
file.
Although very little computation is done by the FPA, it acts as the main conduit for
information between the other five simulation objects. Since multiple FPA’s might be
incorporated into a single imaging system, the Optics and Output simulation objects
are contained within the FPA object and not as separate objects. By directly linking
them to the FPA, IRIMAGE can model multiple FPA’s in a single imaging system
without needing to use the same Optics and Output object for all of the FPA’s. Such
strong links between the FPA and the other objects makes it difficult to alter another
object without affecting the FPA. However, as long as the changes to any of the other
objects do not effect how data is passed or stored, the FPA object should not require
facilitates adding a full featured ray tracer to IRIMAGE. However, for many cases
the simple pin-hole approximation is adequate. Therefore, the Optics object currently
only deals with the pin-hole approximation. Under this approximation, the Optics
object performs three functions. It stores the basic optical parameters that define
the pin-hole approximation. Using these parameters it manipulates Ray objects so
that each ray correctly exits the optical system so that it may be interpolated by
the Scene and Atmosphere VGO’s. Finally, it stores a set of transmission curves
calculation. These three functions define a reasonably simple and accurate optical
system.
a basic optical system. The cardinal points of the optical system define the focal
length and location of the nodal points along the optical axis. Since the WCS should
be defined at the front surface of the optical system, all of these coordinates are
defined with respect to the WCS origin and orientation. Figure 2.14a shows an optical
system with the locations of its cardinal points defined. According to figure 2.14a,
the coordinate of the 2”¢ focal point must coincide with the origin of the FPA object’s
VGO. The focal length used by the Optics object is the effective focal length which
is the distance from the principle point to the corresponding focal point. The Optics
object requires the user to define each cardinal point in terms of its WCS coordinate
along the optical axis. This ensures that the Optics object can then correctly compute
the focal length. Since the optical system is assumed to be optically thick, the user
must also provide the location of the front and back surfaces along the optical axis.
Note that the front surface should be the origin of the WCS, i-e., (0,0,0). Another
parameter to be specified by the user is the optical F-number (F/#). The F/# relates
the diameter of the optical aperture to the focal length (F'/# = D/f). Although most
of these parameters are only used by the Optics object, the FPA object extracts the
focal length, F/# and back nodal point. These parameters are used to setup the Ray
object and convert the spectral radiance to incident spectral power.
a simple change of origin calculation. As mentioned in section 2.2.4, the nodal points
define the points of the optical system where an incident ray striking one nodal point
will appear to exit the other with the same direction vector. Since the FPA simulation
object sets the direction vector of each Ray object to point from the origin of the ray
on the detection element toward the 2Ӣ nodal point of the optical system, only the
Ray object’s origin needs to be adjusted by the Optics object. As one can see from
~ FPA VGO
Effective Focal Le hgth, nar
4 ~
—~ ~ho——
1" Focal 2™ Focal
Point Point
1" Principal Pt. : — 2™ Principal Pt.
(1° Nodal Pt.) Principal (2"" Nodal Pt.)
"Planes"
points, (iii) nodal points. (b)Example of a ray passing through the nodal points of
an optically thick system.
1% nodal point, but at the point it passes through the front surface of the optical
system. Technically, the ray should be refracted at this interface, but the definition
of the nodal point already takes this into account. Therefore, the exit point, Prnit,
can be approximated using the equation (2.4.1) in conjunction with the distance from
the 1* nodal point to the front surface, t, the direction vector of the ray, Rair, and
between the front nodal point and any point on the front surface of the optical sys-
tem. Since IRIMAGE is only interested in the portion of the Ray object that travels
between the optical system and the background scene, the Optics object replaces the
Ray’s origin with the exit point, Purit.
tenuates the incident radiance using its optical transmission curves. Although most
optical systems will only need one transmission curve to simulate the filtering of the
system, IRIMAGE does provide for cases of more than one filter through the use of
multiple transmission curves. As mentioned in section 2.2.4, each curve consists of an
array of spectral values and associated transmission coefficients which form a piece-
wise linear curve. The Optics object linearly interpolates between any two points on
this curve to get the transmission coefficient for a specific wave number. Once all
of the transmission coefficients for a particular wave number are determined, they
are multiplied together to form a single transmission coefficient. Finally, the spec-
tral radiance is multiplied by this transmission coefficient to generate the attenuated
radiance at the given wave number. This process continues for every member of the
spectral radiance array. Once the entire spectral radiance array has been adjusted, it
2.4.5 The Detector Object
ductor, photovoltaic, pyrometer, bolometer. The object-oriented nature of the base
objects used by the Detector object allows all four of the modeled detector types to
use the same core routines for loading detector parameters and computing the output
signal using a responsivity curve. The Detector object maintains a database for each
detector type. Using the detector type and indentifying index value, the database
passes information back and forth between individual detectors and the FPA ob-
ject. This type of organization makes it easy to add detectors to IRIMAGE without
altering any of the other simulation objects.
create an entry for it in one of the four detector databases. Once these databases
are defined, the Detector object cycles through each entry in its parameter file and
loads the proper detector parameters. When the FPA object passes the spectral
power along with a detector’s type and index number, the Detector object uses the
type and index to direct the spectral power array to the proper detector. After it
passes the spectral power to a specific detector, the detector generates the Out Val
data structure which stores all of the data to be output by the detector. The detector
loads the incident spectral power in several forms as well as the signal, noise and
total output from the detector. Once this data has been computed and stored, it is
returned to the FPA object.
DetType is a generic detector object which defines a set of routines to load a set of
standard detector parameters and compute the detector’s output signal using a piece-
wise linear responsivity curve. Det'Type loads all of the common parameters shared by
most major detectors such as the quantum efficiency, fill factor, and integration time
as well as either the detector’s responsivity curve or its D* curve and the data required
to convert it to a responsivity curve. Besides these detector specific parameters,
IRIMAGE, these parameters are transfered by the FPA object along with spectral
power array.
average number of photons incident upon the detector over each spectral bandwidth,
the spectral power must be put in these terms. For the case of MODTRAN, the spec-
tral power provided to the Detector object is in the form of Watts/cm~. Equation
(2.4.2) demonstrates how DetType converts this spectral power at a specific wave
number and bandwidth, P(v)Av, to the number of incident photons at the same
ee bine AV (2.4.2)
form of output /photon. Fortunately, the DetType is flexible enough to provide several
acceptable units for the responsivity curve. As long as the user specifies the unit type,
via a simple integer flag, DetType will convert the responsivity to the necessary form.
Once the conversions have been made, the DetType program computes the output
model defined in section 2.2.6. When these two calculations are complete, DetType
returns the Out Val data structure with these two detector outputs as well as the total
output and the total detectable power.
object. Although most detector types can use the standard DetType routines to load
data and compute the detector output, the object oriented design allows any detector
type to use its own routines. These routines may load more specific data, process
entirely. Two of the four detector types in IRIMAGE utilize this property. For
instance, the photoconductor detector incorporates the generation-recombination (g-
r) rate by processing the computed output. A separate routine multiplies this g-r
rate times the signal, noise and total output. Since the g-r rate is not loaded by the
standard DetType routine, the photoconductor also uses a separate routine to load
the g-r rate. Such modifications do not effect any other simulation objects outside
of the Detector object which simplifies adding new detector types, or improving on
OutVal structure to the Output simulation object for storing in an ASCII file. Like
the Optics object, the Output object is contained inside the FPA object so that each
FPA in a simulation outputs its data to a separate set of files. Since many simulations
may be animated, each frame of the animation is stored as an individual file. For
this reason, each file’s name is made up of a user defined header name and the frame
number. The output data from each detector of the FPA is stored as a single record
in this file. For instance, a 60 frame simulation using a single detector in an 8x8
FPA will have 64 records per file in 60 files. The user has the choice to output any
combination of the following data values stored by a single detector in the OutVal
e Detector type.
e X,Y coordinates for the center of the detection element in the GCS.
e Incident power in watts and/or photons/sec.
e Signal, noise and/or total output voltage/current.
user to store those data values which are important to the current simulation and
discard the rest (saving disk space). Although the Output object currently outputs
only an ASCII file, the object-oriented design allows the detector data to be output in
any form. A future version of IRIMAGE could output images in a standard graphics
format (GIF, TGA, etc.) or even directly to a graphics window.
how the simulator works to produce images. We begin with a brief discussion of the
Graphical User Interface (GUI). The GUI allows the user to define the parameter
files for each of the six simulation objects using a set of intuitive windows and dialog
boxes. Following the GUI discussion, we describe how the IRIMAGE generates an
image from these parameter files. Finally, we present a couple of test results from the
ject’s parameter file. In addition to the six parameter files loaded by the simulation
objects of IRIMAGE, there is a seventh general parameter file. This file stores the file
names and locations of the six other parameter files as well as the general animation
parameters including the starting frame and how many frames to compute. Since
many of the simulation object’s may load multiple incarnations of the same simula-
tion object or load information for other supporting program objects, the parameter
files are divided up into sub-sections which define the parameters for a specific part of
a simulation object. As an example, the GMD has its own sub-section of the Atmo-
sphere object’s parameter file while each geometric model stored in the GMD also has
a sub-section of its own. Such organization makes it possible to add new geometric
models or new detector types without invalidating old parameter files. Furthermore,
to see how a particular simulation works. Despite these advantages, the parameter
files can be very long and involve hundreds of parameters.
in a parameter file was to use a standard editor like emacs or vi. The GUI reduces
the problems encountered when trying to write or edit scripts by hand. We maintain
the flexibility of the parameter file method by using the GUI to generate the files
instead of as a full front-end for IRIMAGE. Since the time to run a simulation may
vary from a few minutes to several days, keeping this methodology makes it possible
to run a simulation in the background without the GUI having to remain open. The
GUI provides a more intuitive interface for the user to load information while making
sure that the parameters are also formatted correctly and placed in the parameter
file in the proper order. In addition, many of the parameters have default values and
the GUI automatically sets them if they are not set by the user. Finally, the GUI can
use old parameter files allowing the user to make small changes in a simulation or
define a series of simulations that share many of the same parameters. Overall, the
the sequence of images desired. To begin, IRIMAGE loads the various parameter files
and initializes all of the simulation objects. Once the parameters are loaded for the
Scene and Atmosphere simulation objects, IRIMAGE initializes each of their VGOs.
Subsequently, it loads the Image, VIO and Layout databases into the Scene object
followed by the GMD, the standard MODTRAN profile and the various gas and
aerosol models into the Atmosphere object. Prior to initializing the imaging system
objects (FPA, Optics, and Output), IRIMAGE sets up the Detector database. After
the FPA simulation object’s parameters are loaded, the corresponding Optics and
Output simulation objects are read in by IRIMAGE. Once the entire imaging system
element to each grid point of the FPA’s VGO using the UnitCell program object as
a guide. As soon as the simulator is initialized, the simulator is prepared to generate
the rays and compute each image.
associated with each detection element of the FPA. Once the rays are all created,
the FPA object transfers them to the Optics object to generate the exit coordinate
and direction vector. Using each modified Ray object, the Scene object intercepts
the ray with the imaging VGO and the Atmosphere object generates the database of
grid point indexes by intercepting the ray with the Atmosphere VGO. As discussed,
these calculations only need to be computed at the beginning of a simulation. For
an animation, this property significantly reduces the work done between successive
frames and improves the speed of the overall simulation. While the amount of memory
required to store the ray’s vector data may be small, the database of grid point indexes
and path lengths requires much more memory. When one considers an FPA with a
large number of detection elements or a large number of rays per detection element,
the memory requirements grow quite large. Therefore, IRIMAGE has two different
memory models to suit various machines running IRIMAGE. If one has sufficient
memory to store all of the rays including each grid point database, the user can specify
that IRIMAGE use the Large memory model. Otherwise, IRIMAGE must use the
Small memory model which preserves the vector data for the entire simulation but
rebuilds each atmospheric grid database just prior to determining the atmospheric
profile and destroys the database as soon as the detector output has been returned
by the Detector object. Both models are built into the simulator so that the user
specifies the model at run time. Once the memory model is chosen and the rays are
constructed and associated with the detection elements of the FPA object, IRIMAGE
can compute each frame.
of detectors, computing the output for each individual detector remains much the
same. The procedure for generating a frame is outlined below as a series of simple
For the small memory model, intercept ray or rays with Atmosphere VGO.
Atmosphere object computes atmospheric profile using Ray object.
Use detection element to assign spectral range to profile.
Pass profile to Atmospheric Object which passes it on to MODTRAN.
to the spectral power collected by the imaging system for the current Ray
For the small memory model, delete Ray object’s grid point database.
6. Repeat step 5 for each element of the FPA.
7. Close the output file.
8. Cycle to next frame and repeat the entire process until all frames are generated.
also clear that the FPA really drives the entire simulator. It acts as an information
server which passes information to be processed between each of the various objects.
The object oriented nature of the entire calculation clearly shows how other steps can
ducted a series of test cases. These cases serve to test both the inner workings of the
simulation (such as testing the capabilities of the VIO’s and the geometric models)
and provide some insight into some of the applications of infrared imaging we intend
Detector Pitch 38 pom
Detectable Range | FWHM=4.610 to 4.776 wm
QEff 65.0%
Integration Time 0.005208 sec
Focal Length 100 mm
place, these examples rely on the proven validity of MODTRAN as an atmospheric
modeling package. Furthermore, we have conducted qualitative experiments similar
to the one in case 2 which provide results that are comparable to the simulation.
IRIMAGE Parameters
First of all, they both use the same imaging system. Table 2.1 points out the major
parameters used by IRIMAGE for the focal plane, detector and optical system. The
imaging system parameters are based on the Amber AE-256 camera system which we
acquired to perform validation experiments.
models to simulate an atmosphere with a polluted region. Both simulations are
based on the 1976 US Standard atmospheric profile which is part of MODTRAN.
The clean atmosphere relies on the gas concentrations of the US standard model and
is at 293K and 1 atm. However, the polluted model increases the concentration of
carbon monoxide (CO) from the standard concentration of approximately 10 ppm to
a much higher concentrations of 1000 ppmv for the ellipsoid and 10,000 ppmv for the
car simulation. The polluted model also is at 1 atm, but its temperature is raised
above the 293K. Therefore, the two simulations we examine utilize the GMD with the
atmospheric VGO to model a hot polluted region surrounded by a clean atmosphere
Scientific Motivation for Simulations
due to the vibrational and rotational quantum modes in the Carbon Monoxide (CO)
gas molecule. The vibrational states place the energy states in the mid-wavelength
IR region (MWIR) while the rotational states produce perturbations around these
vibrational states. The CO molecule absorbs and radiates energy in the range from
4.4 to 4.9 microns (See figure 2.15). Current methods of detecting CO optically utilize
active detection of its absorption signature. Using either a blackbody or laser source,
a spectrometer sweeps across the active region of CO and measures the reduction in
the transmission of the source. However, most sources of CO pollution are hotter
than their surroundings. Therefore, we should be able to measure the CO radiation
instead of its absorption. For this reason, we conducted the following tests with a
heated region of high concentrations of CO.
of high concentrations of CO against a standard background atmosphere and scene.
Using a cigar shaped ellipsoid geometric model that is 50 cm long and 25 cm wide
along the major and minor axis, we associate a region of space with an elevated
temperature of 313K and a concentration of 1000 ppm per volume of CO in addition
to the standard concentrations of the other normal atmospheric constituents. We
place this ellipsoid rotating about its center point at a point 7m from the imaging
system. Outside of this ellipsoidal region, we set the temperature to 293K and utilize
the 1976 Standard Model stored in MODTRAN. At a distance of 10m, we simulate the
background scene using a wall at 293K and emissivity of .9. The simple background
scene prevents mistaking the background for an atmospheric effect and also allows us
to evaluate the capabilities of the atmospheric modeling as well as the detector and
noise model.
The first two images span a voltage range from 0.5 to 1 volts, while the second two
Mode Mode
Total Atmospheric Radiance
(Temperature = 373 K)
a F
9S 5
g 10°F 7
‘pO
(4)
Ey i
8 10’L, 4
a q
& C
5 F == CO - 3.5% mol
m i === CO - Standard
B Wavelength (microns)
of a non-diatomic molecule can either absorb or emit radiation. (b) Radiance of CO
in the 4.0 to 6.0 micron region of the infrared.
noise, Voltage range = 0.5 to 1 volt. (b) With noise, Voltage range = 0.5 to 1 volt.
(c) No noise, Voltage range = 0.73 to 0.755 volts. (d) With noise, Voltage range =
0.73 to 0.755 volts. Bright spots in (c) and (d) are caused by a path that is too short
for MODTRAN to properly compute. They are artifacts of the calculation.
only represent the signal voltage. All four of these images were generated using the
output from a single file and the 2D data imaging capabilities of the AVS imaging
software. Comparing the four images we can see that it easier to resolve a change
from the standard atmosphere to the ellipsoid region than it is to resolve the varying
thickness of the ellipsoidal region. Adjusting the “contrast” by limiting the voltage
range improves the ability to resolve the varying thickness (edges are darker than the
center). However, when we incorporate the noise, the subtle changes in the thickest
part of the ellipsoid become washed out by the noise.
approach as well as the detector modeling capabilities of IRIMAGE. Unlike the sim-
ulations presented in chapter 2 (see figure B.13), we interpolated the ellipsoid onto
a high resolution VGO. This nearly eliminates the “layered” look associated with a
coarsely spaced VGO. Instead, the ellipsoid in figure 2.16c gets darker near the edges
and brighter toward the center, giving it a true three-dimensional appearance. This
is what we expect since the thicker a hot polluted region is, the more it will radiate.
Therefore, this simulation clearly shows that the three-dimensional VGO approach
provides a reasonable method for representing a gaseous region using a primitive ge-
ometry which is connected to a specific atmospheric model. The small voltage range
is also consistant with what we expect from a real imaging system. Furthermore, the
noise model adds significantly to the realism necessary to make adequate judgments
regarding the usefulness of a particular imaging system or application.
ground scene in combination with an atmospheric model. Using the same atmospheric
models as in case 1, this case incorporates an image of a moving car whose surface
temperature ranges from 293K to 305K at a distance of 5m from the imaging sys-
tem. Although fairly unrealistic, we test this car moving across a background image
whose top half consists of a temperature gradient running from 275K to 325K and
CO at 325 K. (a) frame 1, (b) frame 30, (c) frame 60, (d) frame 90.
view of the camera, approximately 2.5m from camera. This simulation is designed to
simulate a car moving across the scene at a distance of 2.5m. More importantly, this
simulation tests the two-dimensional capabilities of the Scene object in conjunction
with an animated atmosphere.
range is higher in this simulation, the voltage range runs from 0.25 to 1.5 volts.
the polluted cone also stands out as well. The varying temperature background affects
the contrast between the cone and the clean atmosphere which could be a problem
when trying to quantitatively detect the concentration of CO in a real plume. By
the final frame, the varying thickness of the cone is slightly discernible although the
temperature gradient of the background image does mask it. The only problem we
encountered was that the atmosphere object must move at a slightly slower rate so
that it does not encroach upon the image of the car. One would expect this result
because we are approximating a three-dimensional scene using a two-dimensional
representation. What actually is happening is that since the car is further away than
the plume, it must travel a greater distance in order to cross the frame. If one keeps
this in mind, it is possible to model many physical situations using this approach.
In case 2, we were able to show that it is possible to combine a 2D background
scene with a 3D atmospheric model to simulate a 3D situation. Like case 1, the slicing
planes are very subtle because we used a much finer grid spacing. This was possible
because the atmosphere VGO only needed to be in the region where the plume would
travel. Futhermore, the concept of the VIO is proven by the overlaying of the car
image on top of the background scene using a simple matte. Finally, the animation
potential to be used as a tool for both application research as well as system design. At
Caltech, we intend to use it to explore the possibilities of passive pollution detection
using multi-spectral methods. IRIMAGE will allow us to figure out what gases can be
detected and problems might be encountered. Furthermore, it will be possible to try
different designs to maximize the detectability of various gases. IRIMAGE will not
only determine what pollution detection methods are possible, but also what systems
effective tool for research and design for now and the future. Despite its many short
comings, IRIMAGE can be easily improved. Throughout its development, many
suggested changes have been incorporated with relative ease. In addition, the objects
were designed with certain improvements in mind such as different noise functions,
new detector models, and more complex geometries. The object-oriented design also
allows future developers to incorporate all or part of IRIMAGE in their own software
which embodies the concept of code reuse.
costly experiments are becoming more difficult to undertake. Tools like IRIMAGE
make it possible to address these experiments from a new direction. Although some-
times more limiting in scope, a simulator like IRIMAGE provides a first cut at a
thermodynamic model and radiation propagation model into infrared three-
dimensional synthetic image generation,” Optical Engineering, 31(7), July 1992,
1505-1516.
ciples approach,” Proc. SPIE, 933, 1988, 199-206.
Gallery, J.E.A. Selby, and S.A. Clough, “Users Guide to LOWTRAN 7, AFGL-
TR-88-0177,” Environmental Research Paper No. 1010, Air Force Geophysics
Laboratory, Optical/Infrared Technology Division, Hanscom AFB, Maryland
(1988).
Hg...Cd,Te. Part I: Status of Stanford University Mercury Cadmium Telluride
Process Simulator,” Journal of Electronic Materials, 24(5), May 1995, 565-572.
lution Model for LOWTRAN7, GL-TR-89-0122,” 1989.
Fox L.S. Rothman, and J.S. Garing, “AFCRL Atmospheric Absorption Line
Parameters Compilation, AFCRL-TR-0096,” 1973.
L.S. Bernstein, D.C. Robertson, P.K. Acharya, L.S. Rothman, J.E.A. Selby,
W.O. Gallery, and S.A. Clough, “The MODTRAN 2/3 Report and LOWTRAN
7 Model,” prepared for PL/GPOS, 1996.
pp. 110.
ter Vapor,” Journal of Meteorology, 12, 1955, pp. 272-284.
pp. 146.
Introduction To Ray Tracing (Academic Press, San Diego, 1989).
Recipes in C, 2nd Ed. (Cambridge University Press, New York, 1992), pp.123-
128.
Oct. 1984, 15-22.
tem,” IEEE CG&A, 5(4), April 1986, 16-26.
Inc., 1994) pp. 35-112.
and the Verification of IRIMAGE
that is simulated and a set of experiments that can verify that the simulation’s output
is logical and realistic. In chapter 2, we discussed the physical background and design
of the IRIMAGE simulator. This chapter discusses the results of the experimen-
tal verification of IRIMAGE. The original motivation behind the IRIMAGE project
was to explore multi-spectral applications of infrared imaging. Since IRIMAGE was
written to explore this topic, our verification experiments not only test the various
imaging capabilities of the simulator, but also serve as the initial investigation into
using multi-spectral imaging in pollution detection. Using a commercial IR imag-
ing system, we demonstrate that IRIMAGE can reasonably and accurately emulate
what an actual imager might see through a series of narrow band filters as well as
over the wider 3 to 5 micron atmospheric window. By comparing the experimental
results with simulations based on the experimental setup, we will show that IRIM-
AGE generates imaging data consistent with these experiments. In addition, we will
demonstrate that multi-spectral imaging, either passive or active, does have promise
for the detection of common gaseous pollutants.
IMAGE should satisfy in order to be considered a valid simulator. These criteria
2. The Atmosphere object reasonably approximates the absorptive and radiative
jects) generates a reasonable and realistic output based on the actual physical
tection using multi-spectral imaging, we designed and conducted three experiments
and their accompanying simulations. The first of these experiments is the Relative
Temperature Experiment (RTE). The RTE images a background scene that con-
tains two thermally isolated surfaces which are at two different temperatures. The
two-dimensional nature of this experiment allows us to verify the Scene object’s ca-
pabilities. The second and third verification experiments both explore using multiple
narrow bandwidth filters to image different atmospheric pollutants. Using a series of
narrow band filters, the second experiment analyzes plumes of warm and cold natu-
ral gas (CH, or methane) exhausted from the standard gas lines entering a typical
laboratory environment. The simulation of this experiment not only verifies the At-
mosphere object, but also helps determine whether the gas line does contain an excess
of COz in addition to the methane. The third experiment utilizes a controlled gas
cell to explore the multi-spectral imaging possibilities of looking at the exhaust of an
internal combustion engine. The gas cell allows us to control the atmospheric content
of a section of the optical path so that we may accurately model the same experiment
using IRIMAGE. More importantly, the gas cell allows us to use hazardous pollutants
like carbon monoxide safely. Since it is impossible to separate the third criterion from
the first two, we use empirical comparisons between the experimental and simulation
images in combination with results quoted by the imaging system manufacturer to
verify the third criterion. After a brief discussion of the imaging system in section 3.2
and some physical background for the multi-spectral pollution detection in section
3.3, each of these experiments is discussed in detail in sections 3.4 through 3.6 of this
3.2 The Imaging System
that are approximately the same as the experimental images, IRIMAGE must model
as much of the actual imaging system as possible. In order to do this, we need
to acquire all of the IRIMAGE imaging system parameters for the camera used in
the experiment. From this information, we can develop a model that reasonably
approximates this camera. However, another set of approximations cause certain
problems with reconciling the images produced by the camera and those produced
by IRIMAGE. These problems can be minimized using a straightforward calibration
process which involves a set of calibration backgrounds. Once this image calibration
system built by Amber Engineering. This system consists of an FPA housed in a
liquid nitrogen dewar, a filter wheel, the external controlling hardware for both the
FPA and the filter wheel. In addition, the actual imaging data can be captured and
stored using a sophisticated 12-bit data acquisition system provided by Amber as well
as a standard Hi-8 VCR. A schematic of this imaging system is provided in figure 3.1.
a dewar and the optical system (See figure 3.2). The focal plane array is made up
of 256x256 individual InSb-based PN-junctions which serve as photovoltaic pixels.
In order to facilitate a rapid set of multi-spectral experiments, the camera utilizes
a cooled filter wheel that can hold up to five filters. Each filter can be rotated
into position either manually or automatically through a computer controlled stepper
motor. Due to the cooling requirements of the FPA and the need to minimize any
thermal effects from the filters, both the filter wheel and the FPA are housed inside a
Controller
Control
[= )@ SHB H&S
Hi-8 Recorder
(Amber-View)
support hardware: computer controllable stepper motor for the filter wheel, FPA
electronics control box, temperature controller (for measuring temperature of FPA),
Hi-8 recorder, video monitor and data acquisition computer.
(CaF2) window provides both an aperture for the cold shield while maintaining the
vacuum that maximizes the cooling efficiency of the dewar. Besides the CaF 2 window,
the camera has an uncooled Germanium (Ge) lens to capture and focus the infrared
radiation. This lens has a focal length of 100 mm, an F/# of 2.6 and a FOV of
5.5°. ‘Table 4.1-1 provides some of the common parameters that describe the imaging
system. These components make it possible to image across the entire 3-5 wm infrared
range as well as several narrower bands within this range.
provides the flexibility to control several key parameters used by the FPA to optimize
data collection. Prior to calibrating the camera for a specific filter and experiment,
Reservoir
Reservoir Electronics
Assembly Cold Finger
where the integration time and the frame rate may need to be varied significantly in
order to improve the image. Once these parameters are set, the FPA can be calibrated
using a two point procedure in which the imager is exposed to homogenous sources
which are at the extremes of the desired temperature range. Using these two images,
the control hardware not only sets the individual pixel gains and offsets to maintain
uniformity across the FPA surface, but it also identifies bad pixels and compensates for
them by substituting neighboring pixel values. After the camera calibration procedure
is complete, the global gain and offset can be adjusted to match the conditions of a
particular experiment. Most of these parameters can be set either from the Amber
electronics control box or via a PC.
as well as in 12-bit digital form. The video may be viewed on a monitor in either 8-bit
Number of Det. 256 by 256
Detector Pitch 38 wm
Detector Type InSb Photovoltaic
Detectable Range 1 wm to 5.5 ym
QE 65
Storage Capacity 1.2710’ charges
Operating Temp. 77K
Eff. Focal Length 100 mm
Optical F/# 2.545
Optical FOV 5.48 °
In our system, we record the black and white signal using a Sony Hi-8 VCR with the
green channel from the control box providing the signal. Using the Amber supplied
data acquisition board and Amber-View software package [1], the 12-bit image data
can be captured and stored. This package also can control the camera via an RS-232
connection so that it serves as both a data acquisition station and a camera control
system. The Amber-View software controls the acquisition of individual image frames
as well as entire sequences of frames. Once these images are captured, they may be
processed, analyzed and stored in several standard and proprietary formats. Initially,
our 12-bit images were stored as 16-bit FITS images [2]. These images were then
processed and stored as 8-bit greyscale GIF images.
are fixed or set by the calibration procedure. Since these parameters are either fixed
or not used by the simulation, there are only a few parameters which are allowed to
vary in each experiment and are simulated by IRIMAGE. These parameters are the
integration time, frame rate of the FPA and the optical transmission curves of the filter
is designed to model all aspects of the imaging system from the optics through to the
output voltage from the FPA. To do so, IRIMAGE uses a set. of parameter files to
define the Optics, FPA, Detector and Output objects. Since IRIMAGE is designed
to simulate various imaging systems, we were able to set up these parameter files to
emulate the Amber AE-256 system. Using the test data provided by Amber for our
focal plane and the optical data for the lens, CaF, window and the various filters,
we constructed a reasonable model of the imaging system. The parameters used to
describe this model in IRIMAGE are provided in table 3.2. A comparison of the
parameters used by IRIMAGE in table 3.2 with the data provided by Amber in table
3.1 shows which of the standard parameters are used by IRIMAGE to simulate the
AE-256 camera. We note that most of the parameters in table 3.1 that are not
used by IRIMAGE are either superfluous or redundant. The two parameters that
are useful but not accounted for are the storage capacity and the effective range.
The storage capacity is directly related to the maximum voltage which the user can
limit when generating the images from the output data. The effective range can be
treated as a bandpass filter in the Optics object or should just be accounted for by the
user when setting the range to calculate. There are other parameters not specified
in either table that need to be accounted for such as the integration time, frame
rate and filter transmission. However, since each filter varies in bandwidth and peak
wavelength, this may affect both the frame rate and the integration time. Therefore,
these parameters are not fixed, but will vary from simulation to simulation. The
combination of the fixed parameters and a few varying parameters allows IRIMAGE
to model the Amber imaging system with reasonable accuracy.
Number of Det. 256 by 256
Detector Pitch 38 pm
Eff. Focal Length 100 mm
Optical F/# 2.545
does not account for bad pixel substitution or the global gain and offsets. Since these
properties may be too difficult to obtain and set (there are over 64,000 individual
pixel offsets and gains) or can be introduced once the image data is generated (post-
processed), we chose not to incorporate these parameters into the current IRIMAGE
modeling scheme. However, the object-oriented nature of IRIMAGE does allow for
the implementation of these parameters at some future time.
object generates an ASCII file with the detector output voltages and various other
pixel specific output (incident power, coordinates, etc.). In order to convert this
ASCII data into a two-dimensional image, we use the AVS Visualization package
from Advanced Visual Systems [3]. Using AVS, we are able to generate images of the
detector output (with or without the background fluctuation noise) or the incident
power. These images are typically 8-bit gray scale images although AVS is capable
of generating 24-bit color images. All of the images generated by IRIMAGE in this
chapter are created using AVS.
individual pixel offset and gain as well as the bad pixel substitution. In addition,
the global offset and gain values also affect the data retrieved by Amber-View. The
analog to digital conversion, the 12-bit nature of the retrieved data and the gain and
offset issues make it extremely difficult to reproduce accurate voltages for each pixel
of the images collected by the Amber FPA. Therefore, direct comparisons between the
actual FPA voltages and voltages generated by IRIMAGE are virtually impossible.
However, comparisons are possible between images generated from the FPA data and
the IRIMAGE data, provided that an image calibration procedure exists to link the
images from the FPA with those from IRIMAGE.
experimental setup which can be reproduced in IRIMAGE. These standard images
utilize two calibration plates. One of the plates is cold (usually room temperature)
and the other is significantly hotter (AT between the plates is 5K to 50K). Since the
Amber camera must be calibrated for each filter using a hot and cold source, these
plates serve as these sources. Prior to the camera calibration, the plates are used to
set the integration time and frame rate. During the camera calibration, images of
the two plates serve as standards to compute the individual pixel gain and offsets to
maximize uniformity across the FPA. In addition, they are used to determine what
bad pixels should be remapped. Once the camera calibration is complete, these two
plates serve another purpose.
ment. By doing so, we are able to adjust the global gain and offset to maximize the
image (voltage) resolution. These two calibration plates define the extreme tempera-
tures for the experiment. With the two plates fixed at their average temperatures, we
use the video image to adjust the global gain and offset of the camera to maximize
the analog to digital resolution between these two temperatures. Since the 12-bit
data returned by the camera has ranges of values between 0 and 4095, we attempt to
set the gain and offset so that the average pixel value of the cold plate is between 200
a plume of 97% CH, at 310.9K against a 298.4K background. Image captured by
looking through a narrow filter centered around 3.454m and bandwidth of .047um.
(a) Range is 0 to 3500, (b) Range is 350 to 550.
background temperature near one of these two extreme temperatures. With the gain
and offset set, we capture an image of each plate, measure the plate temperatures
and store the FPA and temperature data for use with simulation later.
ages into usable 8-bit images. This conversion is necessitated by the fact that most
computer monitors and printing processes are unable to handle more than 8-bits of
greyscale information. Therefore, the experimental images collected by Amber-View
must be converted into a more portable form. Amber-View is capable of setting the
8-bit range of any input 12-bit image to any range with a maximum and minimum
lying between 0 and 4095. It is possible to use this ranging function to increase
the resolution of an 8-bit image by choosing the maxima and minima to be closer
together. This is equivalent to changing the gain and offset of the image. This is
particularly useful for cases like figure 3.3a where the image has subtle differences
that are washed out over the full range. By reducing the range, one can enhance
these differences as in figure 3.3b. We use this capability extensively in the images
experimental images. Using the hot and cold images, we can determine the 12-bit
values that correspond to black and white respectively (black being 0 and white being
255 in the 8-bit representation). Again, we actually stay 5% to 10% above and below
the extremes. This prevents any of the actual experiment images from being clipped
over this range. Once these values are set, we save the hot and cold image data
as 8-bit GIF files and also record the average 12-bit minimum and maximum that
corresponds to that image. Once these two images are saved, the rest of the images are
saved with the same range and subsequently any smaller ranges that might enhance
specific image data. In all cases, one must record the 12-bit minimum and maximum
used to generate each 8-bit image. By doing so, it is possible to compare the data
with other 8-bit images that share the same initial range (between the hot and cold
plate) and the subsequent smaller ranges.
parameters to simulate the hot and cold plates as well as the actual experiment.
Since the average temperature of the plates is known and the plates are painted with
a black paint of known emissivity, it is possible to generate a simulated image of these
two plates. In the case of IRIMAGE, the images are generated from the individual
detector voltages which are floating point values and not 12-bit numbers. We use
AVS to generate an 8-bit image from the voltage values in much the same way that
Amber-View creates 8-bit images from the 12-bit data.
and cold plates, it is possible to calibrate AVS to generate 8-bit images that can be
directly compared with the images captured and processed by Amber-View. Like
Amber-View, AVS allows the user to set the minimum and maximum values that
correspond to black (0) and white (255). By adjusting these values, it is possible to
generate images of the simulated hot and cold plates that have the same 8-bit values
as the images of the experimental plates.
periment using the matched hot and cold plate images from the experiment and
simulation. (a) Hot plate image from experiment. (b) Hot plate image from simula-
tion. (c) Cold plate image from experiment. (d) Cold plate image from simulation.
(e) Image of plume from experiment. (f) Image of cone from simulation.
experimental image and the corresponding simulation image. Methane plume at.
310.9K against a background at 329.2K. (a) Enhanced image of CH, plume from
experiment. (2800 to 3400). (b) Enhanced image of CH, cone representing the plume
from simulation. (467 mV to 555 mV).
for the case of the CH, experiment. We show the hot and cold images for both
the simulated and experimental case of the CH, filter. Since they are nearly the
same except for minor spatial differences, we can then compare an experiment image
directly with the IRIMAGE simulated image and see that they are quite close in
fact. Furthermore, an enhanced experiment image with a smaller range can also be
compared to a similar simulated image provided that the ranges are also comparable.
For smaller ranges, a linear interpolation using the maximum and minimum values
for the hot and cold plates can be conducted. Equation (3.2.1) demonstrates how a
12-bit pixel value (P.op) is converted to an IRIMAGE detector voltage (Prrrarace)
using the experiment image’s 12-bit maximum (Ejaz) and minimum (Fyn) and the
PrrimAGe = E,.,.E,,/ + Vinin (3.2.1)
for generating and comparing experimental and simulated images with a few noted
ing Methods
of emission and/or absorption of electro-magnetic radiation. For the single atom case,
this type of interaction relies entirely on atomic transitions of electrons. In addition to
atomic transitions, many molecules may interact through inter-atomic vibration and
molecular rotation, which both rely on dipole absorption and radiation. Most atomic
transitions require enough energy that the photons must be in the near-IR to UV range
to be generated or absorbed. However, in most cases, the vibrational and rotational
modes of molecules emit and absorb photons of much lower energy and lie almost
exclusively in the infrared region. Since these modes rely on dipole interactions, any
molecules that do not form dipoles (diatomic molecules like No and Oz) will not absorb
or emit photons in the infrared range. Fortunately, most of the gaseous pollutants do
have a significant IR signature while the three primary constituents of our atmosphere
(Nz, O2 and Argon) have relatively insignificant signatures over most of the infrared
range (1 to 100 microns).
atmospheric conditions has been done for more than twenty-five years [4]. How-
ever, most of this work is in the area of terrestrial monitoring from airborne or space
borne imaging spectrometers. An example is the AVIRIS (Airbourne Visible/Infrared
Imaging Spectrometer) project at JPL. AVIRIS uses the reflected blackbody spectrum
from the sun as its source in combination with a “whisk-broom” imager to acquire
224 individual spectral bands between .4 and 2.45 microns [5]. Images are generated
by successive scans as the aircraft moves along the surface of the earth. With the
one axis of the FPA, while the spatial information would be stored along the per-
pendicular axis [6]. Although the same as earlier systems in terms of generating the
image, the spectral data and spatial data is collected simultaneously and images can
be generated much more quickly and reliably. For most earth-bound experiments and
monitoring, this type of system is impractical. Instead, most earth-bound systems
rely on active sources and either full IR spectrometers with a single detector or an
imaging array with a narrow bandwidth filter. Because a majority of these systems
rely on transmission properties of the pollutants, passive detection has been relatively
ignored.
problems associated with trying to image three common gascous pollutants using ei-
ther passive or active multi-spectral imaging. The three pollutants are carbon dioxide
(CO), carbon monoxide (CO) and methane (CH,). These experiments use an imag-
ing system that contains a filter wheel of narrow bandwidth filters with an imaging
FPA. Each filter was chosen to span a region of the infrared spectrum that corre-
sponds to the active spectral region of one of the three gases or a non-active region of
all three gases. The active detection experiments image the transmission of each of
the three pollutants against a constant high temperature background. The passive de-
tection experiments image the emitted radiation of each of three gases at an elevated
temperature against a room temperature background. By collecting and comparing
images of the same background and atmospheric conditions but using different filters,
it is possible to determine whether passive detection is a viable method of imaging
pollution and whether a narrow bandwidth FPA is effective in either active or passive
In general, a molecule with N atoms has 3N degrees of freedom. If we eliminate
the translational motion of the center of mass, the degrees of freedom are reduced to
3N — 3. These other degrees of freedom are made of the vibrational and rotational
states of the molecules. A dipole moment, P, will form if the atoms of the molecule
do not share the electrons equally. Although a molecule may have a dipole moment,
only certain vibrations and rotations will generate the oscillating dipole moment which
makes it possible to absorb or emit photons. For example, in figure 3.6, the carbon
monoxide molecule (CO) has an oscillating dipole for the vibrational state along the
bond between the atoms and the rotational mode around the axis perpendicular to
the bond. However, the other rotational mode around the axis that is along the CO
bond does not cause a change in the dipole moment and thus does not contribute to
the absorption or emission of photons. Experimental evidence and quantum theory
both demonstrate that the effective vibration and rotation modes of a molecule are
quantized. Therefore, only photons with certain energies will be radiated or absorbed
by a molecule with an oscillating dipole moment. In the following paragraphs, we
describe how the vibrational and rotational modes of a diatomic molecule interact to
absorb and radiate photons [7].
Vibrational Modes of a Diatomic Molecule
capable of vibrational motion that is constrained by the bonds between the atoms.
For example, a diatomic molecule only vibrates along the bond between the two
atoms. Figure 3.7 shows a typical energy potential for a vibrating diatomic molecule.
In addition, this figure shows a few of the quantized energy levels for this energy
potential. For the lower energy states, this potential can be approximated by a
quantum mechanical, simple harmonic oscillator (SHO). The ground state energy is
not zero, but is hvg/2 where vp is the fundamental frequency for the vibrational mode
A (P, # P,)
Cc (P i= P. 5]
forms an oscillating dipole. (b) Rotation around perpendicular axis to bond forms
another oscillating dipole. (c) Rotation around bond does not form an oscillating
bond.
curve closely approximates the potential for a simple harmonic oscillator.
1 . .
En = (n+ a)hvo where n is an integer and n >= 0 (3.3.1)
transitions between the different energy levels of the molecule result from the ab-
sorption or radiation of photons of a specific energy. The selection rule for radiative
transitions between vibrational modes (those involving the absorbtion or emission of
a photon) is An = +1. From this selection rule and equation (3.3.1), we deduce
AE, = thir (3.3.2)
dependency in the transition energy. As long as the molecule is not excited to an
energy level outside of the QM-SHO approximation, any transition obeying the selec-
tion rule will generate a photon with frequency, vp. Since the fundamental frequency
for most diatomic molecules lies in the 1 wm to 20 um region of the infrared spectrum,
Rotational Modes of a Diatomic Molecule
Rotational motion of individual molecules is also described quantum mechani-
cally. Specifically, the angular momentum of the molecule about its center of mass is
quantized. This quantum behavior limits the angular momentum to certain allowed
frequencies or “modes.” For a diatomic molecule, the only rotation that contributes
to the dipole oscillation is the perpendicular rotation (See figure 3.6). With this in
L? =r(r +1)h? where r is an integer and r >= 0 (3.3.3)
of a photon obey the same selection rule as orbital angular momentum, i.e., Ar = +1.
Using this selection rule, we can specify the equations for the energy of an emitted or
AE, = + (3.3.5)
_ pnt _ Ar
Vp = EM = (3.3.6)
deep into the far infrared region. Since most IR detectors do not detect radiation at
those ranges, it would be very difficult to image molecules that only radiate through
rotational methods. However, such a case is not physically allowed according to the
selection rules for radiative transitions of a vibrating and rotating molecule. Since
transitions preclude a transition resulting in An = 0 or Ar = 0, all transitions must
the energy required for a transition to take place is just a sum of the transition energies
between a vibration AND a rotation state. This phenomena has been demonstrated
experimentally by the fact that all diatomic infrared signatures do not have a central
peak, but have two peaks on either side which represent the perturbation from the
central peak because of transitions between the ground state (r = 0) and the first
excited state (r = 1) of the rotational mode.
The Vibrational-Rotational Transition Bands of a Diatomic Molecule
From the results above, one can see that the transition energy spectrum for a
molecule with a dipole moment is simply a sum of the transition energies between the
initial vibrational and rotational modes and the final ones. Using equations (3.3.2)
and (3.3.5), we derive the equation for the allowed transition energies of a diatomic
= +h + — (3.3.8)
of this transition determines whether a photon is generated or absorbed. The transi-
tions between adjacent rotational modes splits the vibrational transition energy into a
series of energy bands that are centered around this vibrational energy. As mentioned
earlier, the selection rules disallow the r = 0 energy, preventing a central peak in the
band. Under most conditions, a majority of the rotational transitions occur between
the ground state and the first excited state. Therefore, instead of a single central
peak, there is a peak on either side of the vibrational frequency. Doppler broadening
due to the translational motion of the molecules and the conditions related to the
Heisenberg uncertainty principle cause the individual energy peaks to spread out and
form a series of overlapping energy bands. All of these factors define the characteristic
important to understand the physical difference between active detection and passive
detection. Thermodynamically, if a gas is at a lower temperature than its background,
it will tend toward thermal equilibrium by absorbing the emitted radiation from the
background. On the other hand, if the background is at lower temperature than
the gas, the gas will move toward thermal equilibrium by radiating photons to its
surroundings. Active detection measures the amount of radiation absorbed by gas
from a hot background while passive detection measures the amount of radiation
emitted by the gas to its surroundings.
and emissivity. Using this source as a background, the various pollutants must be
introduced into the atmosphere in front of the background. For a filter encompassing
the active spectral region of a pollutant, the imaging system will detect a measurable
decrease in the incident radiation for regions of the atmosphere with elevated con-
centrations of the pollutant that lie between the background source and the imaging
system. The other filters should have no decrease or a very minor decrease in the
transmission for those same regions of the atmosphere. In other words, a set of images
utilizing active detection will show the transmission characteristics of the atmosphere.
background does act like a source, but only as basis of comparison for the emitted
radiation from the hotter pollutant. In this case, the pollutant is hotter than the
background source, so it will emit radiation which can be detected by the imaging
system. An image, generated using a filter that encompasses the active spectral region
of a pollutant, will show a measurable increase in the incident radiation for regions
with an elevated concentration of the pollutant. The other filters should demonstrate
the same behavior as the active case by showing a minor increase or no increase at
100
2 80
= 60 +
2 40;
= 201
s 0 . . ,
3400 3200 3000 2800 2800 2400 2200 2000
Wavenum ber
Transmission of Carbon Dioxide (CO 2)
100
2 80 +
e OF
2 40
a 20 +
ES Py ;
3400 3200 3000 2800 2600 2400 2200 2000
Wavenum ber
Transmission of Carbon Monoxide (CO)
100
2 80 |
eg 8}
g 40}
Hd
ad 20 |
x 0 ;
3400 3200 3000 2800 2600 2400 2200 2000
Wavenumber
cm7) for (a) CO, (b) COs, (c) CH, [8].
3.3.3 Detection of CH,, CO and CO,
of detecting several common gaseous pollutants using multi-spectral imaging. The
three pollutants we examined were carbon dioxide (CO2), carbon monoxide (CO)
and methane (CH,). Figure 3.8 shows the transmission spectra for these three gases.
Since the vibrational energy is independent of the level transition and the rotational
modes are only perturbations around this energy, the location of the radiative bands
and absorption bands should be approximately the same. Therefore, a filter that
encompasses these bands should allow the FPA to image either the transmission or
emission properties for one of these pollutants. For this reason, we selected a series
of filters that isolate the incident radiation on the FPA to either an active region of a
specific pollutant or a region where all three pollutants should not affect the incident
radiation at all.
passively. In addition, we simulated the experiments using IRIMAGE. We discuss
the rest of the incident radiation. In most cases this is either done through a grating
or narrow band-pass filter. In either case, this reduces the number of incident pho-
tons significantly. Such a reduction will effect an imaging system’s ability to resolve
temperature differences in the background. It is important that IRIMAGE is able
to accurately simulate the imaging effects from different band-pass filters. For this
reason, we conducted the Relative Temperature Experiment which looks at a two
temperature sources through three different bandwidths of the 3-5 um range.
the temperature difference between the standard plate and the isolated region. This
setup can be modeled by IRIMAGE and it verifies the ability of the Scene object
to model simple background sources. In addition, the temperature resolution of the
camera’s FPA for various spectral bandwidths can be measured and compared with
IRIMAGE’s predictions. This also verifies the noise modeling and detector modeling
of IRIMAGE. Finally, the RTE avoids a complicated atmospheric model and uses a
standard model instead. By doing so, we are able to separately test the Scene and
Atmosphere objects.
spectral bandwidths. Using the temperature measurements and filter transmission
curve from these experiments, we simulated each spectral bandwidth and temperature
in IRIMAGE. Using the calibration images, we generated the simulated images and
then compared them with the actual experimental images. By doing so, we were able
the Amber infrared imaging system. The background source consists of an 8” x8” x.25”
sheet of aluminum with a 2.25” diameter hole cut out of the center and an aluminum
disk, that is 2” in diameter, placed inside the hole and surrounded by styrofoam to
thermally isolate it from the larger plate. The small disk has a small contact heater
connected to the back which allows the disk to be heated. The surfaces that face the
imaging system are coated with an aluminum primer and then a standard emissivity
black paint (€ © 0.88). This minimizes any surface reflections from the aluminum. A
thermocouple is attached to the surface of the small disk and the surface of the larger
plate. This thermocouple allows us to accurately measure the temperature difference
between the surface of the small disk and larger area surrounding it. With the heater
2" AE-256
Camera
RTE RTE Source
Source
| Volt
: Meter
ho
the RTE source. To generate the calibration images, the RTE source is replaced by
the calibration plates.
between the two plates. Once the RTE source is ready, it is placed 179 cm away from
the front surface of the optical system and the small disk is centered in the camera’s
field of view. After we focus the optical system on the disk (usually by placing a
finger right in front of the disk), the camera can capture images.
Both plates are 12”x12”x.25” and are painted with the same black paint as the RTE.
The Hot plate has a 12”x12” contact heater attached to its back surface which allows
us to set the plate temperature to values in excess of 325K. The Cold plate has an
attached reservoir which can be filled with ice water to maintain a relatively stable low
temperature plate. In these experiments, the cold plate is kept at room temperature.
These plates serve as the calibration plates for calibrating the Amber camera at the
recalibrate the Amber camera each time a new filter is rotated into the view of the
FPA. Any time the integration time is changed, it invalidates the current camera
calibration. Therefore, after each filter is changed, the following camera calibration
surface. Set the integration time and TIA offset so that the image generated
is bright but not white, i.e., does not saturate the FPA. Repeat with the Cold
plate to make sure that the offset does not cut off the low temperature region.
Write down the values for the integration time, offset and temperatures of the
in front of the lens when prompted by the Amber camera’s software (on the
control box or in Amber-View). This procedure defines the temperature range
of the calibration using the two plates as the two temperature extremes. It also
Generating the Calibration Images
Once a filter is in position and the camera is calibrated, we use the following
the global gain of the FPA until the image captured by Amber-View has a
until the image captured by Amber-View has a minimum pixel value between
3800 and 3900 and the Cold plate image has its minimum between 100 and 200.
For certain filters this is unattainable. In those cases, maximize the difference
and positions of the two plates as well as the image names and the global offset
After capturing the calibration images, the RTE backgound source can be setup
2. Measure and record the temperature of the large plate and the small plate.
4. Heat the small plate until the thermocouple reads the desired temperature dif
5. Record the temperature difference and capture a sequence of 30 frames.
RTE measurement procedures need to be repeated for each temperature difference.
As the temperature difference of the RTE background decreases, the Hot plate tem-
perature should also decrease. By doing so, the gain and offset are set to maximize
the difference between pixel values for the heated disk and pixel values for the room
and the integration time for each experiment conducted are presented in tables 3.3
through 3.5. IRIMAGE uses these parameters as well as those specified in section
3.2 to define the imaging system and the background scene. In addition to these
parameters, IRIMAGE specifies the location of the RTE source (represented by the
Scene object’s imaging VGO) to be (Om, 0m, 2.02m) in the WCS. Since the RTE
source and calibration plates were perpendicular to and centered on the optical axis,
Hot Plate T 337.75 K 311.75 K | 302.80 K
Cold Plate T 297.95 K 297.90 K | 297.75 K
Integration Time | 16.40 ms 16.40 ms | 16.40 ms
Frame Rate 60 Hz 60 Hz 60 Hz
RTE Taise 307.85 K 302.65 K | 298.75 K
(Acentral = 3.319 wm, FWHM = .071 um, Trea = -86).
Hot and Cold calibration plates. The maps are also used for the background plate of
the RTE source. (a) Temperature map, (b) Emissivity map.
Map Map Map
Hot Plate T 336.65 K 312.05 K | 304.55 K
Cold Plate T 297.95 K 297.65 K | 297.75 kK
Integration Time | 5.208 ms 9.208 ms | 5.208 ms
Frame Rate 60 Hz 60 Hz 60 Hz
RTE Tyaise 307.6 K 302.55 K | 298.55 K
(Acentral = 3.930 pm, FWHM = .19 um, Tpeak = -898).
Hot Plate T 334.75 K 327.65 K | 302.85 K
Cold Plate T 297.55 K 297.55 K | 297.95 K
Integration Time | 0.391 ms 0.391 ms | 0.391 ms
Frame Rate 60 Hz 60 Hz 60 Hz
RTE Task 307.15 K 302.35 K | 298.85 K
dow only).
emissivity and matte values using a set of multiplicative and additive constants for
each map. Using this method, the Scene object represents the Hot and Cold plates
using the same image but converted using different multiplicative and additive con-
stants. It also uses this image to represent the RTE source’s large plate. The small
disk is a separate image that contains a matte which allows it to be composited onto
the larger room temperature plate. Figure 3.10 shows the 8-bit images used to define
the temperature and emissivity of the calibration plates as well as the large back-
ground plate of the RTE source. Figure 3.11 shows the temperature, emissivity and
matte images for the small disk which is composited with the background image to
generate the RTE background.
maps did not have a spatially varying temperature (i.e., each map was fixed at a single
temperature). For the RTE experiments, these temperatures were assigned using the
temperature map, the emissivity was varied using an image processing technique of
lightly speckling the surface of the map. This was designed to emulate the physical
painted texture of the calibration and RTE plates. This variance, in conjunction
with the multiplicative and additive constants, causes the individual points on the
map to vary between .88 and .89 (the emissivity of the black paint is specified to be
approximately .88). Finally, this experiment avoids a polluted region and sets the
atmospheric conditions to reflect a standard 1976 U.S. atmosphere at 293K and 1
atm. Using these conditions, IRIMAGE generated a series of images that correspond
with the simulated results. Each figure corresponds to a single temperature difference
(10K, 5K, or 1K) and contain an experimental and simulated image for each filter
used. These figures are organized with the experimental images on the left and the
corresponding simulated images on the right. In addition, a fourth figure is provided
that shows the AT = 1K case with a much narrower pixel value range (and voltage
range for the simulated images). By directly comparing the experiment images and
the simulated ones, it is possible to show IRIMAGE?’s effectiveness as well as some of
its problems.
source for three different temperature ranges. The experimental images and the sim-
ulated images for the LOK difference in figure 3.12 appear to be quite close. Although
a few minor differences are apparent, IRIMAGE successfully images the RTE source
for all bandwidths.
shown in figure 3.13. However, the narrow filter, NO3222, in image A does diverge
more than the other filters. This divergence is reasonable because this filter has a very
c) Wide filter (N03990-4, Pixel Range = 0 to 3900).
(f
(a) Narrow filter (N03322-8, Pixel Range = 0 to 3600).
(b) Narrow filter (N03322-8, Voltage Range = 55mV to 140mV).
(c) Wide filter (N03990-4, Pixel Range = 0 to 3900).
(d) Wide filter (N03990-4, Voltage Range = 293mV to 568mV).
(e) Open filter (none, Pixel Range = 0 to 3700).
(f) Open filter (none, Voltage Range = .745V to 2.08V).
noise may be a contributing factor. In addition, the heated disk is not entirely isolated
from the larger plate and any thermal contact will increase the temperature of the
localized region of the large plate near the disk. None of these factors are accounted
for in the simulation because IRIMAGE assumes only background flucuation noise,
fixes T’ for the two plates and assumes they are completely isolated.
is more pronounced and there is even divergence in image C, filter N03990. Besides
the detector noise discussed above, the randomness in image A may also be partially
systematic. The calibration procedure for the camera relies on a uniform hot and
cold source. Since in this range a minor temperature difference can have a significant
effect, it is possible that the calibration plates are another source of the degredation
in the experimental images. Furthermore, as we reduced AT, we increased the global
offset and gain of the camera. By doing so, we increased the effects of any flawed
calibration. IRIMAGE does not factor in a bad calibration either.
introduced an artifact that was apparent in many of the images. In images 3.15c and
3.15e, we can see a triangular ring pattern surrounding the heated disk. This artifact
appears to be related to the optical system. It changes shape and position when
the filter wheel is rocked back and forth, but it appears for every filter and is more
pronounced with a higher global gain value. In addition to this artifact, it is also
possible to see the initial effects of the noise model in the simulated image. Although
the small disk looks unaffected, the cooler background plate appears to have a slightly
random noise pattern across it. This random pattern appears in all three simulated
images.
the experimental images. The noticeable exceptions appear to occur under conditions
where the approximations in IRIMAGE may begin to breakdown. Other causes of
problems may also include the uncertainty of reading the thermocouple values espe-
cially around 1K when the error is a significant part of the measurement. Also, the
(c) Wide filter (N03990-4, Pixel Range = 0 to 3700).
(a) Narrow filter (N03322-8, Pixel Range = 3500 to 3700).
(c) Wide filter (N03990-4, Pixel Range = 0 to 800).
(e) Open filter (none, Pixel Range = 500 to 1000).
age values generated by IRIMAGE all lie well within the acceptable voltage ranges
specified by Amber for this InSb array (AV % 2.1V). In addition, the size and
physical location of the small heated disk appears to be consistant with the exper-
imental images. Besides demonstrating that IRIMAGE does model the background
scene effectivly, this also demonstrates that the calibration procedure for comparing
spectroscopy results show that most of the common hydrocarbons (methane, butane,
propane, etc.) have significant infrared signatures around 3.4 microns. This wave-
length is associated with the carbon-hydrogen fundamental frequency of vibration.
Since this is the only bond capable of generating a dipole in most hydrocarbons, it
is nearly impossible to discriminate between the different hydrocarbons. However, it
is usually less important to know which hydrocarbon is emitted and more important
to know that they are emitted. Therefore, an imager with a set of narrow band-pass
filters should be able to isolate the hydrocarbon band from another band. By compar-
ing images, it should be possible to determine the existance and rough concentration
of hydrocarbons flowing into the atmosphere from an exhaust source. ITRIMAGE
must be able to simulate what an imager whould “see” under these conditions. The
methane experiment investigates these issues.
highly concentrated plume of natural gas flowing into a standard atmosphere using a
static cone model. Like the RTE, the methane experiments use a simple background.
However, it is not the RTE background source, but just one of the calibration plates
at a specific temperature. In addition, this experiment introduces a small diameter
copper tube that exhausts natural gas from the laboratory gas line. IRIMAGE models
the tube. [RIMAGE represents the plume using the Atmosphere object and two
static cylinders that represent two different concentrations of natural gas. Therefore,
we verify IRIMAGE?’s ability to model localized atmospheric pollution and compare
it with experiment. In addition, we are able to demonstrate how a series of narrow
band filters can isolate emission or transmission bands of various pollutants.
simulations. These filters isolate the CH, band, the CO2 band and a relatively clean
band that lies between the CO) and CH, bands. For each filter, we looked at four
e A Hot plume against the Cold background plate.
e A Cold plume against the Cold background plate.
image parameters. Using the calibration procedure, we are able to generate simulated
similar to the experimental setup for the RTE experiment. The background source
is either the 12”x12” Cold plate or the 12”x12” Hot plate. Both plates are placed at
a distance of 2.43 m from the front surface of the optics. In addition to providing
the background for the methane experiments, the calibration plates again are used
to generate calibration images. These images will be used to calibrate the images
generated by IRIMAGE with ones captured by the AE-256 imaging system.
Natural Gas Camera
Background Plume
by the temperature of water in the tub. The thermal conductivity of the copper
tube and the length of the coil turns the tub-coil combination into a heat exchanger
making it possible to either produce a room temperature or a heated plume of gas.
We measure the temperature of the gas using a commercial gas thermocouple that
has accuracy to the .1K. We also measure the temperature of the exhaust tube and
the background surface using a commercial surface thermocouple capable of the same
accuracy. The 3 mm exhaust tube opening lies approximately 8 cm below the optical
axis of the camera and is located 202 cm from the front of the optical surface.
concentrations of the various gases in the natural gas mixture. The information in
No 0.03
Ox 1.48
CO, 0.22
CH, 92.88
Other CH compounds 5.39
The only constituents we are concerned with are the concentrations of nitrogen,
oxygen, methane and carbon dioxide. Since MODTRAN does not deal with any of
the other CH compounds and they make up less than 6% of the total gas, we neglect
(see section 3.4.2). Although the camera must be recalibrated when the filter changes,
the calibration images do not need to be generated for each of the four cases investi-
gated. Instead, the global offset and gain remain the same for all four cases making
it unnecessary to capture multiple calibration images under the same filter. There-
fore, the image calibration procedure only needs to be conducted after the camera
calibration procedure (See section 3.4.2).
Setting up Hot Plume Cases and Capturing images
ground. The following procedure describes how we setup the experiment and capture
2. Place the coil in a bath of hot water which is approximately 320K.
6. Measure the temperature of the plume, the ambient temperature of the air and
7. Close the gas valve.
10. Repeat steps 4 through 7 with the Cold background plate.
same procedure using a cold plume. For these two cases, the only thing that changes
in the ten steps described above is that the hot bath is replaced with a cold bath.
This cold bath uses water that is approximately 295K. In order to make sure that the
coil is thermally stable at the cold bath temperature, it is necessary to place the coil
in the bath and let it stand for 5 minutes and then replace the cold bath with a fresh
amount of cold water. Other than this minor adjustment, the procedure is exactly
measurements and settings for the four cases investigated under each of the three
filters. These parameters are used by IRIMAGE to model the background and cali-
bration images as well as the parameters necessary to model the imaging system for
each filter. In addition, the coordinates of the background scene (0m, 0m, 2.43m) and
the exit point of coil (Om, -.08m, 2.02m) are already known in the world coordinate
system. Using these values, we are able to place the Scene object’s imaging VGO in
image. Finally, the position of the end of the coil also provides the starting location
Hot Plate T 328.85 K 328.85 K 328.85 K 328.85 K
Cold Plate T 298.65 K 298.65 K 298.65 K 298.65 K
Integration Time | 32.81 ms 32.81ms 32.81 ms 32.81 ms
Frame Rate 30 Hz 30 Hz 30 Hz 30 Hz
Tyas 310.85 K 310.85 K 297.25 K 297.25 K
Trube 311.75 K 311.75 K 297.35 K 297.35 K
Tambient 297.75 K 297.75 K 297.65 K 297.65 K
(Acentrat = 3-319 ym, FWHM = .071 um, Tear = -86).
Hot Plate T 326.75 K 326.75 K 326.75 K 326.75 K
Cold Plate T 297.25 K 297.25 K 297.25 K 297.25 K
Integration Time | 7.813 ms 7.813 ms 7.813 ms 7.813 ms
Frame Rate 30 Hz 30 Hz 30 Hz 30 Hz
Tyas 310.95 K 310.95 K 297.25 K 297.25 K
Tube 311.65 K 311.65 K 297.25 K 297.25 K
Tambient 297.75 K 297.75 K 297.05 K 297.05 K
(Acentral = 3-930 pm, FWHM = .19 wm, Tpeax = -898).
One of these cylinders represents a gas model that is 100% natural gas while the
other cylinder represents a gas model that is only 25% natural gas. Figure 3.17
demonstrates how these two cylinders are placed with respect to each other. The
100% cylinder begins with a radius of 3mm which is approximately the size of the
opening at the end of the coil. At the other end of the 100% cylinder, the radius is
only 1mm wide. Thus, it approximates this region shrinking as the gas moves further
away from the end of the coil. The other cylinder is reversed and has a Imm radius
at the end of the coil and its final radius is 30 mm. Thus, these two objects are a
Hot Plate T 326.85 K | 326.85 Kk 326.85 K 326.85 K
Cold Plate T 297.85 K | 297.85 K 297.85 K 297.85 K
Integration Time | 7.813 ms 7.813 ms 7.813 ms 7.813 ms
Frame Rate 30 Hz 30 Hz 30 Hz 30 Hz
Tyas 309.95 K | 309.95 kK 297.55 K 297.55 K
Trube 311.25 K | 311.25 K 298.05 K 298.05 K
Toambient 297.35 K | 297.35 K 297.05 K 297.45 K
(Acentral = 4.235 ym, FWHM = .181 um, Teak = -87).
the X axis. Their centers both lie at the world coordinate (Om, .036m, 2.02m). We
use an Atmosphere VGO that is centered around the optic axis at the point 202 cm
from the lens and has a grid point resolution of 1 mm in all three directions. In
addition to a standard atmospheric gas model, we also define gas models for 100%
natural gas and 25% natural gas/75% standard atmosphere. The two gas models are
associated with the respective cylinders while the rest of the atmosphere is associated
with the standard atmosphere gas model. The temperatures of these gas models are
all set by the experimental measurements except for the 25% model which is a linear
interpolation between the ambient temperature and the temperature of the 100%
natural gas model.
we model the exhaust end of the copper tube by using the Scene object’s ability to
scale and translate an image and place it in a specific location of the imaging grid.
Like the RTE, the temperature of this image is constant, but the emissivity varies
between .88 and .885. For simplicity, we modeled the end of the coil using the same
image and then scaled and translated it to match the scale and location of the end
of the coil in the experimental images. By doing so, we are able to generate images
that appear to have the plume coming out of the end of the coil. The temperature of
both the background plate and the end of the coil are set according to the recorded
: Natural Gas
exhaust cone that represents the plume.
ground, we report the experimental and simulation results here. Like the RTE dis-
cussion, the figures we present here have the experiment images on the left side and
the simulation images on the right. In this case, we are interested in how IRIMAGE
models the atmosphere versus the actual experimental evidence. It should be noted
that the cone representation used by IRIMAGE becomes less accurate near the top of
the wider portion of the cone. This occurs because the cone model assumes that each
cylinder has a constant concentration. However, physically this is not true because as
the plume expands, the fraction of the pollutant in a plume decreases. Therefore, our
discussions about the experimental plume versus the cone model will center around
the exit point of the nozzle.
For most of the images, the enhancement uses equation (3.2.1) to scale the simula-
tion images to match the experimental images. However, a few images were better
represented by a minor offset in one direction or another. We believe that this offset
COz filter, we used images that were the difference between the captured image and
a standard image taken prior to opening the gas valve. For these images, we set the
voltage range of the simulations so that the scale of the range (AV) matched the scale
of this difference image. We did so by using (3.2.1) and incorporating an additive
constant. In any case, we will note either an offset or difference image in the figure
caption.
is clear that the IRIMAGE cone representation produces a close approximation of
the experimental plume for the CH, filter. Near the nozzle end, they look almost
identical. As the IRIMAGE cone expands, it exhibits the behavior we discussed
above. The background of the simulated image is much brighter for this filter than
the experimental image. This is expected since IRIMAGE does not increase the
methane content in the surrounding atmosphere while it actually does increase in
the experiment. The clean filter appears washed out in the experiment, but a very
faint cone is visible in the simulation. Since this filter does overlap the CO» bands
slightly, we expect that we are seeing just a minor loss of transmission due to this
band. We suspect that the experimental image may also have been saturated by an
integration time too high. Finally, the COz filter appears to have isolated a very faint
carbon dioxide plume. Experimentally, the only way to clearly see this plume was to
subtract a standard frame from the polluted frame. By removing the static clutter
from the background, we are able to clearly see the plume of COg. In the simulated
image, the corresponding cone model is also visible. The bright lines around each
cylinder in the simulation are artifacts of MODTRAN’s computation. Clearly, this
figure demonstrates the atmospheric modeling capabilities of IRIMAGE as well as
demonstrates how an active approach to multi-spectral imaging can be effective.
room temperature background. Again, IRIMAGE seems to accurately reproduce
images that are very close to the experimental evidence (near the nozzle). In this
(a) CH, filter (N03322-8, Pixel Range = 2800 to 3400).
(e) COs filter (N04235-4, Pixel Range = -75 to 25 (difference image)).
(a) CH, filter (N03322-8, Pixel Range = 350 to 550).
(c) Clean filter (N03990-4, Pixel Range = 600 to 700).
simulated image for the clean filter shows the faintest visible cone. Furthermore, the
clean filter in the experiment is not washed out, but still does not appear to exhibit
this plume. Clearly, this figure demonstrates the viability of passive detection of
gaseous pollutants. However, it does appear that it will be more difficult to determine
the concentration of these pollutants. Another important aspect of this figure is that
the noise modeling of IRIMAGE appears to closely match the experimental images
(especially in the case of the COz filter). All three simulated images are definitely
affected by the background fluctuation noise model. Thus, this figure serves as further
evidence of IRIMAGE’s ability to generate accurate representations of the physical
world.
experimental results. Figure 3.20 shows the cold plume of natural gas against a cold
background. IRIMAGE’s simulated cone model predicts that we should see a faint
plume of both CH, and CO,. However, we don’t really see this in the experimental
images. We believe this may be due to slight discrepancies in the temperature mea-
surements of the gas and the background plate. Since the voltage ranges are very
small for all three filters, such minor flucuations could cause the gas to appear in
the simulation, but not in the actual experiment. Furthermore, the noise model does
appear to be less accurate for filters with small bandwidths in the shorter wavelength
regions. Therefore, a signal due to the plume may exist in the experiment, but it is
lost in the noise. Fiqure 3.21 looks very similar to 3.18. In this case, the plume is
cold. However, the results are nearly the same and one can make the same conclusions
as we did for figure 3.18. One interesting note is the minor artifact of the imaging
system in 3.18e. Although not visible in the normal images, this artifact. appears
when we subtract off the standard image. We believe this problem is related to the
calibration for this filter. Overall, these two figures further demonstrate the viability
of multi-spectral pollution detection and validate IRIMAGE’s ability to model such
experiments.
(a) CH, filter (N03322-8, Pixel Range = 325 to 425).
(c) Clean filter (N03990-4, Pixel Range = 600 to 700).
(a) CH, filter (N03322-8, Pixel Range = 2800 to 3400).
(e) COg filter (N04235-4, Pixel Range = -75 to 25 (difference image)).
modeling generate reasonable results. However, we believe that the noise model does
need some improvement. Our assumption has always been that the FPA operated
under BLIP conditions. For filters at shorter wavelengths with small bandwidths,
the detector noise most probably dominates causing the simulated images to look
“cleaner.” A further improvement to IRIMAGE will be to incorporate detector and
readout noise models. More importantly, the results of this experiment clearly demon-
strate that multi-spectral imaging, both passive and simple active, show promise and
and carbon dioxide (CO2). According to Seinfeld, an internal combustion engine
operating under normal conditions is capable of producing 0 to 10 percent CO and
4 to 14 percent CO, in a dry air mixture [9]. Figure 3.22 shows roughly how the
concentration of these two pollutants are related with respect to air-fuel ratio. The
concentration of CO goes down linearly with respect to air-fuel ratio while the CO,
concentration increases. The use of a catalytic converter reduces the concentration of
CO further to meet most state and federal standards. The 1986 federal standards for
CO are 3.4 g/mi for light-duty automobiles and 17 g/mi for light and medium-duty
trucks [10]. In the course of our experiments, we discovered that Linde Specialty Gas
company provides a series of specialty gases that are used to calibrate automobile
emissions testing equipment. The lowest concentration of both gases is a mixture of
1.6% CO and 10% CO, [11]. The intent of the Gas Cell Experiment is to emulate
a car exhaust plume (for an idling vehicle) on a small scale and demonstrate that
it is possible to measure CO and CO, independently using a set of small narrow
band-pass filters. At the same time, the experiments also provide a real world test
of the abilities of IRIMAGE to accurately generate images that are comparable with
12
by
~—
y 10
o «8
Re)
4 6
> 4
Pd
o 1 l 1 i li L\ L L
8 rt) 12 14 16 18
Air-Fuel Ratio
engine. [9]
gas cell that would contain the exhaust plume of CO and CO: and prevent the gas
from mixing with the atmosphere of the laboratory. For each experiment, we purged
the cell with dry nitrogen to create a clean atmosphere. Once the cell was filled with
a dry nitrogen atmosphere, we introduced hot and cold plumes of carbon monoxide,
carbon dioxide and a mixture of both against either a hot or cold background. Using
the lessons learned from the methane experiments, this cell uses the same copper
tubing with the water bath to generate a heated plume. The GCE also uses the
calibration plates as the backgrounds for these experiments. Using methods similar
to those in the methane simulations, IRIMAGE models the atmospheric conditions
using two cylinders. These cylinders are linked to atmosphere models which define
different temperatures and concentrations of CO and CO». Besides further verifying
IRIMAGE’s abilities, this experiment investigates the viability of imaging a polluted
plume using an imaging system equipped with several narrow bandwidth filters.
Nozzle Baffle AE-256
Camera
Background Gas Cell li
Manifold
e A plume of 1% CO and 99% No.
e A plume of 1% CO, 6.6% CO, and 92.4% N» (automobile emission).
and a hot plume against a cold background. Unfortunately, certain conditions regard-
ing our gas cell windows made it difficult to generate decent images in any of these
cases. Therefore, in order to compare the simulated images to a set of experiments,
we used a simple subtractive image processing technique to improve the experimental
setup except for the introduction of the gas cell. The background source is either the
‘.
: 7
‘7
Chris Springfield & Wes Saizillo :
California institute of Technology an: 5
a distance of 2.83 meters from the imaging system. As in the previous experiments,
these plates serve as both the background sources and the calibration sources.
ardous pollutants like carbon monoxide. The gas cell is a glass tube that is 50 cm
long and 43.8 cm in diameter with walls that are 3/8” thick (see figure 3.24). There
are five ports on the cell. The four outer ports can be used for exhausting the gas
inside the gas cell to the outside (we used three of them and plugged the fourth).
The middle port is for the gas input tube and the gas thermocouple. The gas input
tube is the same copper tube used in the methane experiment, submerged in a water
bath to regulate the temperature of the gas flow. Since there is long portion of the
tube between where it exits the water bath and where it ends inside the gas cell, we
applied a surface heater to the end of the tube to prevent the heated gas from cooling
before it enters the gas cell. All of the ports are sealed by rubber stoppers which
have pass through holes for the exhaust or input tubes. In addition, the stopper in
the middle port has a second hole for passing the gas thermocouple into the gas cell.
Using the gas thermocouple, we are able to measure the ambient temperature of the
Gas Cell Manifold
Valve Output
Vent
Valve
High Rate | @© @© 6@ : Mixing
N, Manifold
manifold contains four individual flow meters which have separate inputs but have
a common output through the mixing manifold. In order to form our three plumes
and purge the polluted atmosphere between experiments, these individual flow meters
are connected to a cylinder of dry No, a cylinder of dry COz and a cylinder of 3%
CO/97% No. The No, COg and the CO/N»2 mixture are each connected to separate
flow meters; the Ng is connected to a high-rate flow meter for purging the gas cell
and a low-rate flow meter for use in mixing the other two gases to the appropriate
concentrations. By controlling the flow of each of the gases into the manifold, we can
accurately set the plume’s exit concentration.
windows back into the camera, the ends of the gas cell were ground to an angle
of 5°. Initially, the windows were made of 1/16 inch thick Plexiglas. Plexiglas is
approximately 88% transmissive over the 3-5 micron range and has the strength to
avoid flexing during an experiment. Unfortunately, the camera was unable to see a
this reason, we replaced the Plexiglas windows with a polyethelyne-based plastic wrap
stretched over the two ends of the gas cell. With these makeshift windows, we were
able to see the CO2 plume and keep a tight seal to prevent the gas from leaking out. of
the gas cell. Although more transmissive, these new windows introduced another set
of problems. Because the plastic wrap is so thin and capable of stretching and flexing,
it exhibits a certain amount of non-uniform transmission. In addition, varations of
the pressure in the cell caused the windows to bow in and out, resulting in unwanted
reflections of the cell and the end of the copper tube.
which would minimize the reflections for any surrounding objects (lab bench, camera
case, other lab equipment, etc.). As shown in figure 3.23, the baffle is a cardboard
box which is painted black on the inside and contains a pyramid shaped set of walls
inside. This shape prevents most direct radiation and reflections from entering the
optical system and thus reduces the unwanted reflections to a much smaller number
experiments (See section 3.5). For this experiment, we imaged the CO2 plume through
all three filters before moving on to the CO and the mixture plumes. After we
completed the COz plume experiments, we decided that it was unnecessary and less
accurate to examine only one type of plume at a time. Therefore, the CO and the
CO/CO, mixture experiments were both done for the various backgrounds and gas
temperatures before changing the filters. In order to minimize the difference between
the COQ experiments and the other two experiments, we made every effort to duplicate
the conditions for the camera and image calibration.
We did not capture different calibration images for each different background or plume
concentration. Instead, we maintained the same global gain and offset for every
experiment conducted using a specific filter. Like the camera calibration, we made
every effort to duplicate the image calibration conditions of the CO, experiment with
the conditions for the CO and gas mixture experiments.
Purging Gas Cell
1. Check to make sure the CO, and CO cylinders are closed.
2. Make sure water bath is filled with cold water (= 15°C).
3. Close the valve to the gas cell and open the direct exhaust valve.
4. Purge the manifold with Nz at a rate of 7 1/min for 1 minute.
5. Open the gas cell valve and close the direct exhaust valve.
6. Purge the gas cell at the same Nz flow rate for 20-30 minutes.
7. Check for CO/COz levels using combustion analyzer.
8. Continue purge until CO is under 15 ppm and COz is under 1%.
Imaging a Hot Plume Against a Cold Background
plume against a cold background is the same. The following procedure describes how
1. Flow Ng at approximately 2 1/min through gas cell.
Close the valve between the gas cell and the manifold.
3.6.3. Simulation Setup
background temperature measurements and gas parameters for each of the three
filters. Since several parameters are the same for the hot plume/cold background
and cold plume/hot background cases, we placed the parameters for both cases in
the same column of these tables. Using these parameters, IRIMAGE models the
background and calibration images as well as the atmospheric conditions and the
imaging system for each filter. The coordinates of the imaging VGO are (Om, 0m,
2.83 m) in the WCS while the end of the copper tube in the gas cell is at an angle
of 6 degrees below the horizontal (X-axis) and is at the WCS coordinates (-.127m,
09m, 2.464m). Although the tube is barely visible in the experimental data, we have
Hot Plate T 319.45 K | 319.55 K 319.55 K
Cold Plate T 294.65 K | 294.75 K 294.75 K
Integration Time 5.21 ms 5.21 ms 5.21 ms
Frame Rate 30 Hz 30 Hz 30 Hz
H/C: T source 319.45 K | 319.55 K 320.25 K
A/C: Vyas 294.65 K | 294.15 K 294.60 K
H/C: Tambient 293.65 K | 294.00 K 294.45 K
C/E: T source 294.65 K | 294.05 K 293.85 K
C/H: T gas 299.95 K | 301.95 K 301.95 K
C/E: Tambient 295.95 K | 295.25 K 295.40 K
(Acentral = 4-235 pm, FWHM = .181 pum, Tpeak = .870)
(C/H: Cold Background, Hot Plume - H/C: Hot Background, Cold Plume)
a mixture of both using the same two cylinder construct as in the methane experiment.
The small cylinder contains 100% of original concentration of the pollutant (CO,
COz2 or mixture) while the larger cylinder is reduced to 25% of the original pollutant
Hot Plate T 316.45 K 316.75 K 316.75 K
Cold Plate T 293.35 K 294.70 K 294.70 K
Integration Time | 32.81 ms 32.81 ms 32.81 ms
Frame Rate 30 Hz 30 Hz 30 Hz
H/C: T source 316.45 K 316.75 K 316.80 K
H/C: Vyas 294.05 K 294.95 K 295.00 K
H/C: Tambient 293.55 K 294.95 K 295.05 K
C/H: Tsource 293.35 K 295.50 K 295.00 K
C/H: Teas 300.95 K 301.45 K 301.25 K
C/H: Tambient 293.85 K 295.65 K 295.65 K
(Acentrat = 3-686 pum, FWHM = .042 yum, Tpeak = .727)
(C! Cold Background, Hot Plume - H/C: Hot Background, Cold Plume)
Hot Plate T 318.75 K | 318.15 K 318.15 K
Cold Plate T 294.65 K | 294.25 K 294.25 K
Integration Time | 5.21 ms 5.21 ms 5.21 ms
Frame Rate 30 Hz 30 Hz 30 Hz
H/C: T source 318.75 K | 318.15 kK 318.15 K
H/C: Tas 295.25 K | 294.75 K 294.65 K
H/C: Vambient 295.35 K | 294.25 K 295.75 K
C/E: T source 293.75 K | 293.75 kK 293.95 K
C/H: Vyas 302.15 K | 303.35 K 302.05 K
C/H: Tambient 294.25 K | 296.85 K 295.45 K
(Acentrat = 4.693 wm, FWHM = .167 pm, Tpear = -785)
(C/H: Cold Background, Hot Plume - H/C: Hot Background, Cold Plume)
nozzle having a radius of 4mm while the other end has a radius of 2mm. The larger
cylinder is also 25cm long, but the radius nearest to the nozzle is only 2mm while
the other end has a radius of 30mm. We approximate the orientation of the exhaust
cone by rotating these two cylinders so that their centerlines lie in the XY plane of
the WCS and are rotated around the Z axis at an angle of 6°. The center point of
both cylinders is placed at (.0lm, .085m, 2.464m) in the WCS.
calibration plates (See figure 3.10). Since we are not modeling the end of the exhaust
tube in this case, the Scene object only loads a single image. This image either
represents the hot background plate or the cold background plate. Each simulation
sets the plate temperature based on the experimentally measured temperature while
the emissivity is set to lie between .88 and .885.
This would require modeling a spatially variant filter outside of the imaging system.
Unfortunately, IRIMAGE can only simulate a partially transmissive surface on the
imaging VGO outside of the imaging system (Scene object using a soft matte). There-
fore, these problems must be ignored in the simulations. This factor must be kept in
carbon dioxide and carbon monoxide using passive multi-spectral IR imaging. We
present the results of the experiments and subsequent simulations here. We used a
three filter setup so that we could isolate the CO and CO, bands from each other.
The third filter is a clean filter which allows us to discriminate the background sources
from the plume. For each plume type (CO and CO» concentration) and filter, we only
looked at the two cases of a hot plume against a cold background and cold plume
against a hot background. These two cases represent the extremes that one might
encounter when passively imaging a plume of polluted gas. Since the validity of
IRIMAGE has been demonstrated by the previous two experiments, the comparisons
discuss the simulated results as predictions of what the expected images should be.
This discussion will demonstrate that passive imaging of these pollutants is possible
but does face certain problems.
detecting the plumes. This made comparing experiment images with simulated im-
the NO-4235 filter. Demonstrates the problems caused by the gas cell windows and
shows the image processing solution.
(e) Expt: Diff. Range = -25 to 175. (f) Sim: Diff. Range = 689.2mV to 706.6mV.
starting point failed to generate images that clearly showed the plumes. However,
using a standard image processing technique of subtracting a standard image (no
polluted plume) from these images, we were able to significantly reduce the effects of
the windows. Figure 3.26 shows the experimental and simulated images of a plume
of 10% CQOz over the full range, a narrow range and the image processed range of
values. Although the IRIMAGE cone model predicts that. the plume should be visible
in all three ranges, the problems with the windows mask out any noticeable change
in the full and narrow range images. But the image processed version does show the
expected CO, plume. In most cases, the expected plumes became easier to see in the
image processed images. Some of these plumes were only visible when watching a se-
ries of frames and seeing changes in the area of the image that we expected the plume.
Therefore, in this discussion, there may be images where a plume is expected but not
noticeable. For those cases, we will point out if a plume was visible when looking
at series of frames. All of the experimental images presented have been processed
by this subtraction technique. The simulated images use a voltage range consistant
with the range of the experimental image and an offset to make the simulated image
match the overall appearance of the experimental image. In all of these figures, the
experimental images are on the left and the simulated images are on the right.
Figures 3.27 through 3.29 present the images of cold versions of the three plume
types against a hot background. By looking at the results of all three experiments
together, we can make some simple observations about each experiment and also
compare them together. In figure 3.27, the plume consists of 10% CO», and 90%
Ny. We can see that IRIMAGE predicts that the plume will only be visible through
the COz filter. The images from the experiment agree with this prediction. A clear
plume of carbon dioxide does appear to be flowing out of the nozzle in figure 3.27c,
but nothing seems to be visible through the other two filters. For the case of a
plume of 1% CO in figure 3.28, IRIMAGE predicts that we should see a plume of CO
that is fainter than the CO, plume. We also note that the MODTRAN artifacts in
3.28d may indicate that the CO band is slightly visible through the CO, filter. An
) Clean filter (N03689-4H, Pixel Range = 0 to 150).
(f) CO filter (N04693-4, Voltage Range = 1.519V to 1.557V).
(a) Clean filter (N03689-4H, Pixel Range = -150 to 0).
) Clean filter (N03689-4H, Pixel Range = -75 to 75).
(f) CO filter (N04693-4, Voltage Range = 1.486V to 1.540V).
plume is not visible through the CO filter. However, this occurs because the image
processing is unable to eliminate the effects of the windows without washing out the
plume. By watching a set of 30 captured frames, the CO plume becomes noticeable.
Unfortunately, a single image is not enough for this case. As predicted, the plume is
not visible through the other two filters. As a result of these first two experiments
and the subsequent simulations, it is clear that the CO and CO, filters are capable of
isolating the two consituents from each other while the clean filter prevents any part
of the background source from being mistaken as part of the plume.
to determine the existence and relative concentrations of the two gases based on the
above results. The experimental and simulation results presented in figure 3.29 are
from an example of such a case. In this experiment, the plume consists of nitrogen
with 6.67% CO, and 1% CO. IRIMAGE predicts that both a plume of CO, and
CO should be visible. Experimentally, the results are less clear. A faint CO2 plume
is visible through the COz filter, but again the CO plume is not visible in a static
image. Like before, the CO plume is only visible when watching an animation of the
30 captured frames. The clean filter does remain clean, which confirms this plume is
not part of the background. We conclude that for the case of a hot plume against
a cold background, it is possible to determine the existence of both CO and CO,
using multi-spectral methods. However, further experimental work must be done to
determine the ability to accurately measure the concentration for cases like this.
plume containing CO, CO: or a mixture of both against a cold background. While
the previous case relied on the absorption of the background radiation by the plume,
this case relies on the emission of radiation by the CO and CO, in the various plumes.
Before discussing these three experiments, it must be noted that a reflection of the
nozzle is visible for all three filters and should be ignored (one of the reasons for the
clean filter). In addition, the edge of the gas cell and other reflections are visible in
(a) Clean filter (N03689-4H, Pixel Range = -50 to 50).
(a) Clean filter (N03689-4H, Pixel Range = -100 to 100).
(a) Clean filter (N03689-4H, Pixel Range = -100 to 50).
that a bright plume of CO, will be visible through the COs filter, but neither of the
other two filters. The experimental results agree. The image looking through the CO,
filter does have a bright plume against the cold background. The other two filters
show no trace of the plume. In the case of the plume containing 1% CO and 99%
nitrogen, IRIMAGE predicts that the plume should be visible through the CO filter.
In addition, IRIMAGE predicts the presence of a very faint plume through the CO,
filter. This confirms our earlier observation that the CO band is wide enough to be
detected by this filter. However, the signal is very faint making it likely that it may
be lost in the noise or overwhelmed by the existance of COg. IRIMAGE does predict
the clean filter to remain clean. Unlike the other case, the plume of CO is visible in
the static image and is not visible through the other two filters. Like the results from
the hot background/cold plume case, it is still possible to isolate the CO and CO,
bands using the two filters. In addition, it appears that it is easier to detect a plume
of CO under these conditions than in the other case.
6.67% CO., and 1% CO against a cold background. In figure 3.32, the simulations
predict that the CO and CO2 plumes will both be present and detectable. Similar to
the case of the hot background and cold plume, the experimental results agree with
what we expect. A plume of CO is visible through the CO filter and a plume of CO,
is visible through the CO, filter while the clean filter remains clean. These results
in conjunction with the results from the other cases prove that pollution detection
using passive multi-spectral infrared imaging is possible under a variety of conditions.
Although possible, these experiments also point out that futher experimentation and
simulation is necessary before a system should be built. Another important result
of these experiments is the successful use of IRIMAGE as a supplemental tool for
3.7 Conclusions
frared imaging. The experiments conclusively verify that IRIMAGE is capable of
simulating the infrared imaging process with reasonable accuracy. They also point
out certain areas that should be addressed in future versions of IRIMAGE. Although
this verification is important, the results related to pollution detection are even more
significant. These experiments clearly demonstrate the viability of gaseous pollution
detection using passive, multi-spectral infrared imaging. In addition, it is clear that
further experimentation and simulation leading to certain advances in technology and
system design will also be necessary. Overall, these experiments were quite successful
in achieving the goals set forth and should serve as the starting point for further
investigation into multi-spectral IR imaging.
were comparable to images generated by experiment. Although certain discrepancies
existed, most of these were related to the calibration procedure and certain approx-
imations made in representing the gas plumes and the surfaces. The calibration
procedure relied on a constant temperature across the entire plate surface. Unfortu-
nately, the methods employed to heat the surfaces, could not successfully maintain
such a situation. In any case, the effects were minor. Although the approxima-
tions were made when the simulation was setup and not inside IRIMAGE, they were
neccessary because of certain limitations in IRIMAGE. For instance, we could not
properly approximate the plumes using a many cylinders because MODTRAN re-
quires the individual layers of the profile to be greater than a 1mm to avoid artifacts.
Adding more cylinders would have caused more artifacts which would have made it
more difficult to tell the difference between a correct calculation and an artifact In
addition, an acceptable plume model would have required additional code to generate
a properly diluted plume that did not incur the same problems with MODTRAN .
Until a better atmospheric model is available, this problem will persist.
the filter bandwidths got smaller and their peak transmissions became less than 4
microns, the noise model began to under-estimate the amount of noise in the image.
This is most likely due to the detector noise becoming the dominant noise in these
cases. For the cases of short wavelength and small bandwidth, the number of incident
photons approaches, and may go below, the minimum number of photons required
for background limited performance (BLIP). When this occurs, the detector noise
begins to dominate (definition of the BLIP limit). Therefore, IRIMAGE needs to
incorporate detector and readout noise for filters with short wavelengths and small
bandwidths.
multi-spectral infrared imaging in pollution detection. It was possible to isolate and
detect the constituents of several different plumes using a multi-filter arrangement.
Although we were able to qualitatively detect the plumes, quantitative determinations
of the gas concentrations may be more difficult. This difficulty lies in the fact that
the detected change in the incident photons due to a polluted plume relies on the
temperature of the background source. For a hot background, the plume reduces the
number of photons through absorption. For a cold background, the plume increases
the number of photons by radiating them. A solution might be to use the third
filter (or more) as a temperature sensor. If we could determine the temperature and
emissivity of the background, it might be possible to process the images from the two
filters to account for the background source. Further experimentation and simulation
should be conducted to determine the best method of solving this problem.
recalibrate the camera for each filter made it impossible to conduct these experiments
in real time. The recalibration process took between 15 and 30 minutes each time.
In order for multi-spectral IR imaging to be useful, this calibration time must be
on the order of a few milliseconds or eliminated entirely. Using current technology,
this might be possible using filters that have central peaks and bandwidths that do
average number of photons collected by each filter must be of the same order. If
each filter had similar transmission curves, the bandwidths of the shorter wavelength
filters would need to be larger than the longer wavelength filters. Furthermore, the
temperature range of the measurement would greatly influence the filter selection.
Although a system like this would be reasonable in an academic or research setting,
the expense and maintenance of such a system makes it impractical as a commercial
system. Therefore, a new type of imaging system is needed such as an imager with a
micro filter on each detector; a stack of detectors, each with a different cutoff; or even
multiple FPA systems with a different filter per FPA. The challenge is to determine
the best systems for passive, multi-spectral infrared imaging and implement it, and
1992).
(Nasa/OSSA Office of Standards and Technology, Greenbelt, MD, 1991).
More information about AVS can be obtained from: Advanced Visual Systems
G. Vane and A.F.H. Goetz, “Terrestrial Imaging Spectroscopy,” Remote Sensing
of Environment, 24, 1988, 1-29.
Porter, “The Airbourne Visible/Infrared Imaging Spectrometer (AVIRIS),” Re-
mote Sensing of Environment, 44, 1993, 127-143.
for Eos,” IEEE Transactions on Geoscience and Remote Sensing, 27(2), March
1989, 136-144.
third ed. (Aldrich Chemical Co, Milwaukee, WS, 1981).
(John Wiley & Sons, New York, 1986), p. 93.
[10] J.H. Seinfeld, Atmospheric Chemistry and Physics of Air Pollution
(John Wiley & Sons, New York, 1986), p. 101.
Volume 25, p. 106.
IRIMAGE
block diagrams describe how the major programming objects are organized. They
are designed to give the reader a graphical view of how the various pieces of IRIMAGE
are put together to form the major elements of the simulation. The following is a
Database and the Imaging VGO are contained within the Scene Object.
and the Atmosphere VGO are contained within the Atmosphere Object.
and Unit Cell Database are connected through the FPA Object.
that are part of larger databases or part of elements of a database (like key frame
databases) as diamonds; and non-database elements of a major database as well as
objects that store important data as circles. These diagrams do not attempt to
Simulation
di ¥
Scene Detector
Object Object
¥v
Atmosphere
Object
WV
FPA
Database
FPA FPA
#1 #N
Object Object Object Object
#1 #1 #N #N
Object
y Imaging
VGO
Layout
Database
VIO
Database
Image
Database
Object
Object
Model
DB
Model
Database
Cone Box
DB DB
Cylinder Eliipsoid
DB DB
Layout
Database
VGO
Object
Detector Detector
#1 #N
Detector Detector
#1 FN
#N
#1
Object
Output
| Object
Optics
Object
Unit Cell
Database
Filter
Database
#1
Database
#N
DB Array DB Array
Data onto a 3D Orthogonal Grid
environment. The most straightforward approach is to define geometric models as
bounding surfaces and assign each model a certain location and orientation [1]. This
method can produce amazingly photo-realistic images and has become the most popu-
lar form for computer graphics. Another method creates primitive volumetric models
(spheres, boxes, etc.) and places them in the 3-D environment to form more complex
models. Modeling techniques like Constructive Solid Geometry [2] and voxel-based
octrees [3] utilize this method. Volumetric representations may not produce photo-
realistic images, but they are quite useful for imaging applications which require faster
image production or contain lots of volumetric data. The third technique employs
some form of mesh generation algorithm, such as the Advancing Front method [4, 5],
to represent the surface of an object by filling the space surrounding the object with a
3D mesh of points. We call this technique Negative Space Filling (NSF) because the
interior of objects is not filled with the mesh, only the empty space around the objects.
This technique is useful when solving three-dimensional differential equations numer-
ically or conducting a finite element analysis (e.g., aerospace applications, mechanical
engineering, etc.).
curate surface representation, but it does contain important volumetric information
about the models in the scene. The most popular form of this type of data rep-
resentation utilizes the concept of voxels. Voxels are box structures which act like
representation of a geometric model. The simplest form of a voxel representation is
called Space Occupancy Enumeration (SOE) |6]. An SOE representation divides the
space surrounding and containing a geometric model into a regular set of 1-bit boxes.
Each box is either inside the object (bit = 1) or outside (bit = 0).
voxels that make up the geometric model are stored; the rest are discarded. Such a
procedure only represents individual geometric models and not entire scenes. We have
chosen to modify this SOE-Voxel approach by using a concept from the NSF method.
In our case, we divide our entire working space into a regular lattice of identical grid
boxes, i.e., voxels, to form an orthogonal, three-dimensional grid. We define, store
and manipulate this grid and the individual voxel information through the use of
the Versatile Grid Object (VGO) which is discussed in section B.2. Unlike the NSF
method, we do not remove points and/or adjust the lattice to define bounding surface.
Instead, each voxel stores a specific set of values that associate it with a particular
geometric model or the space between models. As one can see, this extends the SOE
representation beyond the simple 1-bit methodology.
an object-oriented construct which facilitates the placement of volumetric data into
the individual voxels of the VGO. In section B.3, we introduce the concept of the
Universal Primitive Object (UPO). Each UPO represents a specific set of volumetric
data using a geometric model that defines the bounding surface of the data and
an accompanying data index which links a set of data to the volume enclosed by
the surface. The geometric model employs a polygonal-based surface representation
which approximates a three-dimensional surface with a patchwork of two-dimensional
polygons. The UPO is flexible enough to build virtually any primitive object while
it is also well suited for our “Slice and Dice” interpolation algorithm.
based models onto a VGO. This algorithm employs a set of cutting planes that slice
each UPO along the Z-axis of the VGO. The result is a series of curves on the cutting
plane that represent the intersection of the plane with the surface of the UPO. These
tion algorithm. This 2D grid is a projection of the 3D VGO on the cutting plane and
represents the voxels that intercept the cutting plane. Finally, our algorithm asso-
ciates a voxel with the volumetric data of the UPO if its associated 2D grid rectangle
has more than 50% or its area inside the intercepted curve. In this way, bounded
surfaces are interpolated onto the VGO.
algoithms are used in the application IRIMAGE. IRIMAGE uses the concepts devel-
oped in this paper to model the conditions of the atmosphere in an infrared scene.
This application takes advantage of the various aspects of these program objects and
or floating point data in a 1, 2, or 3 dimensional, regularly spaced, orthogonal lattice
construct, ie., grid. The grid contains its own coordinate system and transformation
matrices which link any point in the Grid Coordinate System (GCS) to a point in
World Coordinate System (WCS). Grid point data entry and retrieval can be con-
ducted using either a set of three integer indexes (X,Y, Z) or real coordinates (z, y, z)
in either the GCS or WCS. In addition, support exists for using 3-Vector representa-
tion of these coordinates or indexes (7 = xX + y¥ + 2%). Furthermore, support exists
to assign a specific “object index” to grid points that contain data from a particular
object (image, UPO, etc.). These properties make it possible to store, manipulate
and retreive data associated with sources like images and three-dimensional bounded
tem (GCS). The origin of the GCS lies at the center of the VGO’s grid, defining a
& A
Translation
Vector
4 y
bn Be 4 iy
z » Gy o » Gy
gf vas ras
hiss ris
“WY \y
7" a> Wy Wy
Gx Gx
Wx Wx
A Versatile Grid Object in WCS B Translation
Gz
Wz
a 7
i)
a. ya a AY
IY
"1 B y
see > Gy
ae,
Va
“bz
p—> Wy
Gx
Cc Rotation no Change of Scale
(b) Translation in Wy direction. (c) Rotation about Gx axis. (d) Change of Scale
adjusts grid point spacing.
will use the VGO have their own coordinate system, we define the overall coordinate
system of an application to be the World Coordinate System (WCS). Given the lo-
cation and orientation of the VGO’s grid in the WCS, the GCS constructs a set of
transformation matrices that link the GCS to the WCS. These matrices map a point
in the WCS to a point in the GCS and vice versa by translating and/or rotating the
point. The only requirement for the GCS, with regards to the WCS, is that the GCS
and WCS are based on the same units. In other words, the distance between any two
points in the WCS is the same after they are transformed to the GCS. The spacing
between each grid point is defined with respect to the GCS so that each grid point’s
location is defined in the GCS. The physical size of the VGO may be scaled, but this
does not affect the transformation matrices because it only scales the spacing between
grid points and not the GCS itself. Figure B.1 shows the relationship between the
GCS and WCS and how various transformations effect this relationship. Note in this
figure that change of scale only makes each grid box larger, but does not affect the
coordinate system. This relationship between the GCS and the WCS allows all op-
erations internal to the VGO to be completely disconnected from the WCS, thereby
simplifying many of these VGO operations.
formation matrices. These matrices correspond to the standard form for three-
_ 10004, — |o s, 0 0
T(d) = S(s) =
The Rotation Matrices (R,(@), Ry(w), Rz(¢))
0 cos? —siné 0 0 1 0) 0
Rx (9) = Ry(w) =
0 sin@ cos@ 0 —sinw 0 cosw 0
[0 0 0 1 I 0 0 O 1]
cose —sing 0 0|
sing cos6 0 0
R,(¢) =
0 0 1 0
verse transformation matrices. The order of this multiplication is extremely important
Mecs—wes = T™*- R,* -Ry*- Rzt (A.2.2)
G=S8-G (A.2.3)
As mentioned, a change in scale only effects translating the GCS coordinates of the
grid point into the integer indexes required for retrieving data from the grid. In
addition, the scaling matrix is always stored and acts on the spacing vector whenever
the spacing vector is called. Thus, the original spacing remains unchanged so that
Index Array (OIA) and a one-dimensional Flexible Data Array (FDA) that stores ei-
ther integer or floating point data. The OIA is an array of indexes that link particular
data set objects (images, UPO’s, etc.) with the grid points that are affected by the
object. For instance, if a geometric model like a box or a sphere is interpolated onto
the VGO, each grid point residing inside the model would have its OIA index set
to the unique integer that represents the model. This makes it possible to quickly
determine which points are associated with a particular object when changing the
VGO or removing that object from the VGO. If this link is unnecessary, the OLA
Variable Indexes ——<_—=p
oe Ria 3b ~ a °
3 | 8 FI Fl cl ct mic
t (ig) = (2,N-1) t= N1 | | t= 2n-1 | t= 3N-1 |
stored in a 2D data aray. (b) Actual 1D array structure.
a VGO with N grid points and M data values stored at each point, the FDA stores
a data index. The one-dimensional nature makes it possible to dynamically allocate
the memory at run-time instead of fixing one or more of the dimension sizes as required
for a multi-dimensional array. The variable order and number of variables may be
defined at anytime including during run time. There are several easy to use routines
for loading and retrieving both the OIA index and the various FDA values. This
protects the integrity of the VGO, while allowing the programmer to manipulate the
VGO’s contents. The VGO also maintains a default array to store a set of default
values for each grid point. This array is useful whenever the program needs to return
a grid point to a default setting.
a specific grid point location is defined by three integer indexes corresponding to the
three primary axes of the GCS. In addition, a fourth index determines which one of
variables stored at each grid point to access. The integer indexes are always positive
where the (0,0,0) position represents the minimum physical coordinate of the VGO
and the (Ximax,; Ymar, Zmax) Tespresents the maximum physical coordinate. Since the
VGO’s grid is orthogonal and the grid points are regularly spaced, any lattice point
can be defined by a simple integer triplet (X,Y, Z) which does not need to be stored.
Instead, the regular nature of the VGO allows it to use equations (A.2.4) and (A.2.5)
to convert a multi-dimensional integer index into a single unique index for accessing
the one-dimensional arrays.
X,Y,Z = Integer index values for the current grid point
coordinate, (x,y,z), in the GCS. Using the quantities defined above with the grid
to the nearest integer value.
data for a grid point by using its WCS coordinate. By transforming the coordinate
to GCS, it can then use the above schemes to retrieve the nearest point in the grid
and its associated values. In addition, the accessing scheme may be used for 1D and
2D cases by setting the number of points along one or two of the axes to a value of
one, e.g., Nz = 1 for a 2D grid. The accessing scheme then adjusts instantly.
or other objects onto it. Instead, each of these objects that define 2-D images and
3-D bounding surfaces must have their own interpolation algorithms. By doing so,
the VGO is separated from any of the objects that might be interpolated on it. This
avoids the VGO being constantly adjusted for new geometric objects and interpolation
B.3 Defining Bounded Volumes for Interpolating
volume to be interpolated on the VGO. The two major components of a bounding
volume are the bounding surface which defines the physical extent of the volume and
the parameters that make the volume different from its surroundings. As discussed,
various methods exist for defining the bounding surface of an object. Polygon-based
surface modeling is accepted to be the most flexible and straightforward method-
ology for generating geometric surfaces. With this in mind, we introduce a hybrid
form of polygon surface modeling that improves the interpolating of surface models
on the VGO. In addition to the surface modeling technique, it is also necessary to
incorporate the volumetric data into our approach to modeling bounded volumes. To
accomplish this, each bounded volume is represented by a three-dimensional, “smart”
geometric model called a Universal Primitive Object (UPO). We use the term “uni-
versal” because it is capable of building any type of simple primitive surface model
(sphere, box, cone, etc.) as well as many other more complex objects. Each UPO not
only contains the algorithms and variables ncecessary to build the surface model, but
Modeling
volume, we surveyed the methods for creating geometric models in computer graph-
ics (CG). The most prevalent method for defining models is to build the bounding
surface from a set of two-dimensional polygons wrapped around the surface. This
produces a piece-wise approximation of the surface. Furthermore, all other surface
representations can be approximated using this method. Since this method is the
polating three-dimensional volumes onto the VGO.
polygon-based method treats and stores each polygon of the bounding surface as a
separate geometric object. Each polygon stores its own point coordinates and is
independent of all other polygons. As can be seen in figure 3-la, every point that
is shared by more than one polygon is stored multiple times. Although this is the
quickest method to generate a surface, it is highly inefficient.
define its circumference [8]. In addition, each edge end point on the surface is kept
in a separate point array. By doing so, each edge just stores the point array indexes
of the two end points in the edge array. Since the edges are also stored in an array,
each polygon just stores a list of array indexes as well. Figure B.3b demonstrates how
the polygons are inherently linked to each other through their common edges. This
method does require a few more steps to generate the model, but there is a significant
improvement in memory use.
plane intersection may require cycling through the same edge twice. However, the
edge-based method will only need to intercept any edge once, making it more efficient
in both memory and speed. In our algorithm, we utilize a hybrid approach which is
basically the edge-based method with a few minor additions. This algorithm utilizes
three distinct data structures:
point on the surface using a 3-Vector data type. The array structure is just an
unordered one-dimensional array of 3-Vectors. The array can be unordered since the
points will be linked to edges via their array index.
Point Point Point Point
Surface Representation
stored in both the Polygon A
and the Polygon B
Edge Pair Arrays.
ID
#1
#2.
2 3 Polygon Edge Edge Edge
ID #1 #2 #3
c 3 1 A a b c
d 1 4 B c d e
e 4 3
Surface Representation
Edge ID, but not the two
points that make up the
polygon). (b) In an edge-based system, each edge that is common to two polygons is
discussed later.
The Polygon Array Structure
edges that make up the polygon. We use triangular polygons which automatically
maintains an ordered list of edges. In any polygon surface model, it is best to use tri-
angular polygons when possible. Although less efficient memory wise, using triangular
polygons ensures that each polygon is always planar (required for correct modeling)
idea of a Universal Primitive Object was developed. The UPO has the same structure
and generation properties of the classic primitive shapes (ellipse, cone, cylinder and
box). In addition, the UPO’s structure allows defining and modeling of extruded
shapes and surfaces of rotation. Although it uses the edge-based concepts, the UPO
can easily be extended to the classic polygon-based method as well. In either case,
each surface is built using triangular polygons. While the UPO may be slightly
memory inefficient, it is these characteristics that allow the universal idea as well as
improve the speed of the building and interpolation process.
end points. The series of control contours define the external point structure of the
UPO. The contours can define most of the surface, but they can’t close the surface
at the two ends of the surface. Instead, the UPO caps off the surface by connecting
the first end point to the first control contour and the last end point to the last
control contour. Both the control contours and the end points must be defined in the
UPO’s own coordinate system which is called the Primitive Object Coordinate System
(POCS). This coordinate system does not need to coincide with the WCS, but like
in the POCS can be transformed to either the WCS or the GCS.
Control Contours
same number of points. Each contour is defined using either a pre-defined set of points
or is built using a center point, radius value and orientation vector. Most primitive
shapes may be defined by contours which are built using a center point, radius and
orientation vector. However, more complex shapes may require the contours to be
pre-defined. The only restriction is that only one of these methods may be used to
define a particular UPO.
or both of the end points. If the contours are built, the points are regularly spaced
around the contour by the angle @. In addition, every other contour is rotated by 6/2
with respect to its adjacent contours. This provides a smoother surface as well as a
more generalized method of connecting multiple rings.
nects the individual points of each contour to form the edges of a closed contour. Then
the adjacent contours (or end points) are connected to the current contour to form
the edges of a set of triangles that wrap around the surface. Figure B.4 demonstrates
how an ellipsoidal object is created using this method. The concept of creating edge
connections between points on each contour and adjacent contours is described as
connected to the point Pj4,. In addition, P; of Rp is also connected to points P; and
models because every contour is out of phase with its adjacent contours by theta/2.
This phase difference makes smoother looking curved surfaces, but can’t make a
simple box. However, eliminating the phase difference also has problems because if
Control Contours
P)
P) 6
ry) Os
0 @
Cc D
control contours and end points of the Ellipsoid. (b) The first contour’s edges are
connected and the first end point is connected to the points on the first contour to
form the top cap of the Ellipsoid. (c) The rest of the contour edges are connected.
Adjacent contours are connected using the method of connecting point P; on one
contour to the points P;_; and P;,, on the adjacent contours. (d) The last contour
and the second end point are connected to form the bottom cap.
Contours
es 6. (Rotated by 6/2)
id Og
go eo”
_ Q
Unconnected
Box: (a) The control contours and end points of the Box. Note middle contour is
smaller than the outer two contours and the end points lie in the same planes as
the first and last rings. (b) The first contour’s edges are connected and the first. end
point is connected to the points on the first contour to form the top face of the Box.
(c) The back contour’s edges are connected, but the middle one remains unconnected.
Connect point P; on the middle contour to points P;_; and P,,, on the other two
contours to form one set of polygons. (d) Connect point P; on the first. contour to
point P; on the contour ring to form the other polygons (with two edges from the
previous edge creation step).
to the other points. The second method requires that the points along every other
contour remain unconnected (i-e., points along the contour path are not connected).
In order to build triangular polygons, each connected contour is connected to both
of the adjacent unconnected contours as well the next pair of connected contours.
This forms two distinct forms of triangles. One set are long and are formed by
corresponding points on two connected contours and one point on the unconnected
contour. The other set of triangles are formed by two points on a connected contour
and one on the unconnected contour. This method has one restriction; the UPO must
use an odd number of control rings (3 or more). This method produces extruded
surfaces and prism objects accurately. Figure B.5 demonstrates how a rectangular
box can be built using this method. The general principle for creating geometric
and the points along contours Rn, and Rp; are connected. Point P; on Ry is
connected to points P; and P41, on Ryyy and Rp_y. In addition, P; on Ry 18
tie all of the points of the first and last contours to a central end point to form a cap
for the surface. This preserves the desired triangular polygon requirement. As an
added benefit, it is easier to define certain objects like cylinders with cone tops and
bi-pyramidal models.
e Number of points per control contour (N,t).
e Number of control contours (Nping).
e Flag whether control contours are to be built or are pre-defined.
2. Pointer to Pre-defined Contour Parameters (stored in contour order).
e Flag whether to alternate contour connections or not.
e Location of two end points.
Algorithm for Building UPO
1. Create Point, Edge and Polygon Arrays.
e Create Edge Array (Array Size = 3NpingNpt)-
2. Load UPO Point Array
e Determine angle between points on each contour: 6 = 27/Np:
point.
— Create temporary point array for contour n.
— Determine extra half angle rotation: a = 0(1 ~ mod(n, 2))/2
— Use unit radius to compute points in temporary XY plane:
P,(t) = cos@;, Py (i) = sind; where ¢; = i0 +a
Prinat(j) =T,- P"'(i) where j = nNp + 7.
(b) Contours Pre-defined by User
e Load pre-defined array of points.
3. Build First Cap of UPO (Edge and Polygon Relationships).
e Build first contour’s edges.
e Build top polygons from the three edges.
4. Build Multiple Contour Connections.
— First contour: second polygon defined now.
(b) Method 2 - Alternating between connected and unconnected contours.
~ Adjacent diagonals do not have a vertical edge between them.
— Both connected contours form polygons with diagonals.
e Begin to build vertical polygons with vertical edge and one diagonal.
— Two of three edges of polygon are defined.
e Finish building vertical polygons with the corresponding diagonal edge.
e Build edges for second connected contour.
e Associate edges with polygons
— First connected contour: second polygon defined now.
— Diagonal and Vertical edges: both polygons defined.
~ Second connected contour: first polygon defined now.
5. Build Second Cap of UPO.
e Build polygons from two adjacent edges to end point and contour edge.
only has to define a few general parameters. A couple of examples are given in table
B.1.
Contour Radius R, = R3 , Rp = V2- Ri | Use Equation with Nring & Nope
information makes translation, scaling and rotation a simple exercise of transforming
the points; the integrity of the model is maintained.
specific UPO. This data is stored either as an array of floating point or integer values
or as a single integer index which links the UPO to another structure that contains
the data. In either case, the UPO simply must load and store this data. The main
surface with a three-dimensional, orthogonal grid works adequately to load the vol-
umetric character of a bounded volume while maintaining a vozel approximation of
the original surface. Imagine dividing a three-dimensional block of space into a set
of regular rectangular boxes called voxels. Each voxel has the same dimensions and
represents the bounding box that surrounds an individual grid point. All properties
of the voxel are stored by this grid point. Let a series of volumes be placed into this
grid space, with the restriction that their volume must ultimately be represented by
these voxels. Therefore, each grid point contained inside one of these volumes would
be loaded with the characteristics of the bounded volume. Figure B.6 shows an ex-
ample of a cone being interpolated into a three-dimensional grid. The final result is a
i.e., in figure B.6, they are opaque and not transparent.
as indexes to other data sets, various gas or chemical concentrations, temperature,
absorption, etc. All of the uninterpolated voxels would contain a default form of
these characteristics. If an algorithm were to “drill” into or “ray trace” through the
grid, it could quickly create a profile of these characteristics along the direction that
UPO-based Geometric Model
which are perpendicular to the Z axis of the VGO. Each of these cutting planes is
spaced so that it represents a single plane of voxels in the XY plane. In other words,
each layer of voxels along the Z axis has its own associated cutting plane. Figure
B.7 shows that the cutting planes pass through the center of each layer of voxels
so that the planes coincide with the locations of the actual grid points. Once the
cutting planes are defined, the bounded surface is intercepted with the plane. This
plane-surface interception may create a single curve or a series of curves in the slicing
plane. If the surface is bounded, these curves will be closed. If the surface is also
non-intersecting (i.e., surface does not intersect itself), then the curves will also be
non-intersecting. The closed curves can then be intercepted with the two-dimensional
layer of voxels which lie in the cutting plane. Since the Universal Primitive Object is
bounded and non-intersecting, it is ideal for use with this algorithm.
advantage of the additional link between any edge and the two polygons that share
the edge. Most. methods of intercepting a plane and a polygon surface model involve
cycling through each edge to determine which edges are intercepted. Although ex-
cellent for hidden surface removal or getting interception points, if you are interested
frame of a 5x5x5 VGO. (b) A cone as the bounded volume to be interpolated. (c) The
cone placed inside the VGO. (d) The final interpolation of the cone onto the VGO.
problems because of this additional link between each edge and the polygons that
share it.
placed and oriented in the WCS, it can be transformed into the GCS of the proper
VGO. Since the UPO’s point array is the only place where the physical coordinates
are stored, only this point array is affected by the transformation to the GCS. Once
in the GCS, a plane-surface interception is reduced to cutting the UPO with a set
of parallel XY planes. Furthermore, the UPO remains in the GCS and all of the
interpolation algorithms are conducted in the VGO’s GCS. Once the interpolation is
complete, the UPO is returned to the WCS and may be reused again.
through the edge array to find all of the edges intercepted by the plane. The GCS
in a data structure. In addition, the two polygons that share the intercepted edge are
also stored in the same data structure. Once all of the edges are found, the algorithm
cycles through all of the intercepted points and uses the indexes to the stored polygons
to determine how to order the interception points into a set of piece-wise continuous
curves.
loops through the other points until it finds another point that shares one of the
two polygons associated with the first point. Once the second point is found, then
the algorithm uses the polygon not shared by the first two points to find the third
point. At the same time, it stores the index to the other polygon associated with
the first point as the end polygon index. The algorithm continues the cycle until it
finds a point that has the same other polygon index as the end polygon index. This
establishes that this point is the final point of the curve and that the curve is closed.
Once the curve is closed, the points associated with the curve are stored in a separate
array and removed from the intercepted point array. The algorithm continues until
there are no more points left in the intercepted point array. The final result is one or
grid has already been developed [9]. However, this algorithm deals strictly with
unordered curves intercepted with a 2D grid. It does not consider any relationship
between the curves intercepted on the grid. However, there may be times when the
slicing operation may result in multiple closed curves being inside one another (see
figure B.8). Not only must the curve be interpolated on the grid, but the order of the
curves interpolated must also be considered. Otherwise, the final result might not be
what is desired (see figure B.9). Therefore, the slicing algorithm includes a method
for sorting the curves from the outer most curve to the inner most one. Then, the
grid interception algorithm.
There are various methods to sort a series of curves. Our algorithm uses the
piece-wise continuous nature of the curves. The piece-wise nature allows us to define
Count the number of times the ray strikes each curve. If the ray strikes a curve
an odd number of times, then the line segment is inside the closed curve. If the
ray strikes the curve an even number of times, then the curve does not contain
are piece-wise continuous and non-intersecting, then the above method can be used to
sort the curves. Since each curve is made up of line segments and is non-intersecting,
one can choose any line segment that makes up part of the curve. Using this method,
the interpolation algorithm determines which curves contain the line segment and,
subsequently, the curve. Given that none of the curves intersect, i.e., each curve either
lies completely within or outside another curve, the algorithm orders each curve by
the number of other curves that it lies within. For example, if a curve lies within two
curves, it is a level two curve. If a curve does not lie within any curves, it is a level
zero curve. Thus, the curves can be sorted from the outer most curves (level 0) to
passing through Torus to form top and bottom. (b) Result of slice in A is one curve
inside the other. (c) Cutting plane passing through torus to form two unconnected
pieces. (d) Result of slice in (c) is two independent curves.
Torus. (b) A correctly interpolated Torus. (c) An incorrectly interpolated Torus (no
internal ring).
Selected Line Segment
A a set of Curves B the Line Segment Level
Letter Result
ce | ¥3)
D |N(2)
c Sum up Curve Interceptions D for entire curves
curve. (a) Example of a line segment inside a series of non-intersecting curves. (b) Ex-
tend ray along one direction of line segment. (c) Count the number of intersections.
If odd number, then line segment is inside curve. If even, the line segment is not
inside curve.. (d) A curve containing the line segment also is inside the same number
of curves if all of the curves are non-intersecting.
figure 8b). (a) Two curves overlaid on grid. (b) Outer curve interpolated first places
volume data in the grid (An even level # is additive). (c) Second curve overlayed on
interpolated grid demonstrates what grid points should be returned to default value.
(d) Both curves interpolated on the Grid.
dimensional grid determines the area of each grid rectangle that is contained inside
this curve. If the area is greater than 50%, then the associated voxel is marked as
being inside the object. Otherwise, the voxel remains unchanged. If a solid is in-
tercepted and it creates multiple curves on the cutting plane, then these curves may
represent holes or parts that are not connected to the object in the cutting plane.
The level number of the curves determines which curves will encompass voxels and
which curves will return changed voxels to their original state. For instance in figure
B.11, the outer most curve, level 0, encompasses every grid rectangle inside the curve.
However, the level 1 curves define holes in the solid and, consequently, subtract the
effected area from those grid boxes contained within both level 1 and level 0. Fur-
thermore, a level 2 curve would define areas in these holes which are again part of
the solid. In summary, all of the even numbered levels are treated normally by the
algorithm and any area of the grid rectangles inside these curves is added to the ef-
fected area of each of these grid rectangles. All of the odd number levels return the
area inside the curve to normal by subtracting the effected area from each effected
grid rectangle. In this way, the curves are “diced” and volumetric data is interpolated
on the VGO. Once an entire UPO-based surface is interpolated on the VGO, it is no
2. Compute bounding box for object in GCS.
y §
edges.
e Store indexes to two polygons shared by edge.
Store intercepted areas for each grid rectangle in temporary grid.
e Odd numbered levels subtract from intercepted areas.
e Pass any other values associated with UPO surface model to
VGO grid point.
B. If less than 50% of the area is inside the UPO, skip the grid rect-
IRIMAGE Simulator
(IR). Polluted atmospheres generate specific IR. signatures which may be detected us-
ing IR detectors and focal plane arrays (FPA). The modeling of such atmospheres and
imaging systems should make it possible to determine what conditions are necessary
to passively image various gaseous pollutants in the atmosphere. In order to model
an atmosphere, one must be able to define the conditions everywhere in the viewable
area. One way is to use a three-dimensional grid structure which can approximate
these conditions by cutting up space into smaller boxes filled with a homogeneous
gas.
to approximate a clean atmosphere that has regions of higher pollution concentration.
In the simulator, all or part of the atmosphere is represented by a three-dimensional
VGO. Any polluted region is represented by a bounded volume in the form of a
UPO. Different atmospheric conditions are stored in the form of two databases: gas
concentration models and aerosol models. Therefore, each UPO is associated with a
particular atmospheric model through the database indexes of a gas model and an
aerosol model. In addition to these two index values, the UPO also stores its own
object index. Subsequently, the VGO initializes the OIA to store the object index
voxel with a pair of default gas and aerosol model indexes which represent the initial
atmospheric conditions. As the simulation proceeds, various UPO’s are interpolated
on the atmosphere VGO using the interpolation method specified in section 4. IRIM-
AGE then ray traces through the VGO to determine an atmospheric profile along the
ray. The profile is passed along with a value for the background scene temperature
and emissivity to the atmospheric modeling algorithm, MODTRAN [10]. In MOD-
TRAN, the final radiance for the background scene and atmosphere is computed and
returned to the imaging system. Since most atmospheric systems are dynamic, the
simulation utilizes a “key frame” animation extension to the UPO. The rest of the
section details how IRIMAGE takes advantage of the VGO and UPO when computing
profile observed by each element of the simulation’s Focal Plane Array (FPA). The
advantage of using a ray tracing technique with the VGO is the relative ease of
intercepting a ray with a three-dimensional, orthogonal grid and generating a gas
and aerosol model profile along that ray. In addition, the ray also intercepts a two-
dimensional background scene (a 2D imaging VGO behind the atmosphere VGO) to
get the temperature and emissivity of the background. Most ray tracing applications
utilize a ray tracer to incorporate accurate reflective and refractive effects to an image.
However, the voxel nature of the VGO prohibits using this method to approximate
any reflective or refractive effects. Overall, the greatest. advantage of the Ray-VGO
interception methodology is the fact that. the VGO remains fixed over time and just
the contents of the VGO change in time. Therefore, each ray only needs to be traced
once if the ray stores each intercepted voxel’s grid point index in an ordered stack
of indexes. In addition, it must also store the path length of the ray through each
voxel. With this information stored in an ordered stack, the ray generates a profile
the voxels changes because a new UPO has been interpolated on the VGO, the ray
simply rebuilds the profile using the new state of the voxels along its path.
not the actual gas and aerosol models stored by each grid point. Therefore, rays
can be built and intercepted with the VGO before loading the UPO-based bounded
volumes onto the VGO. Each ray is defined by a starting point, a direction vector
and an end point (which is where the ray strikes the imaging VGO of the background
scene). Using this information, each ray is intercepted with the VGO. The inter-
ception method uses a simple plane-line intersection algorithm that takes advantage
of the uniform box nature of the voxels in the VGO. Each voxel is treated as the
interception of six planes. Thus, the entire VGO’s grid structure may be defined by
extending planes that are perpendicular to each axis and are spaced according to the
scaling vector, G, of the VGO. Once the VGO is defined in this manner, then the
ray-VGO interception reduces to a series of ray-plane interceptions. Each ray-plane
interception point is stored temporarily, but in the correct spatial order in which the
ray strikes each plane. Once the ray either passes entirely through the atmosphere
VGO or strikes the imaging VGO, the algorithm cycles through the ray-plane inter-
ception points to determine which voxels were intercepted. As it cycles, it computes
the midpoint between adjacent interception points. It passes the coordinate of the
midpoint to the VGO which returns the nearest grid point index. This establishes
that this portion of the ray passes through the grid point’s voxel. Using these same
two interception points, the algorithm computes the distance between adjacent, points
and stores that as the length of the ray in that particular voxel. Once all of the grid
point indexes and path lengths are determined, the ray is capable of generating an
atmospheric profile (see figure B.12).
UPO-based volumes can be intercepted with the VGO. The position, orientation and
scale of the volumes in the WCS are controlled by the KeyFrame program object. This
program object stores a set of “key frames” which define specific spatial conditions
in Atmospheric Grid and Stored in Atm. Grid
eo e oe e ® e e 8 e
e e es @ @ e
e eo S
e es o
es e e
oe o e
es e eo
e e@ eo
Ray traced through Grid Data Interpolated
Atmospheric Grid and Stored in Ray
e @ e e 2 eo e ° so e e | e @ e e °
eo ° e 2 e 2 e os 2 e @
efotes efe efe « ee
re] re |
e G@ioeheie
e efefede
o E ° e oo os
2 i i efeiele
ul
Temperature & Emissivity
From Background Scene Grid
Convert Ray Data | Profile Processed
into an a By MODTRAN
Atmospheric Profile | to Generate Radiance
we i
Four Layer i
Atmospheric Profile u
Cc
atmospheres are interpolated onto the VGO. (b) Ray passes through the VGO and
generates the stack of grid points that it intercepts. Then it retrieves the current
atmospheric indexes for the profile. (c) IRIMAGE converts stack of values into an
atmospheric profile. MODTRAN converts this atmospheric profile into an array of
radiance values to be input back in the imaging system.
the user only needs to determine size, location and orientation of the UPO for certain
frames. In addition to the animation of individual UPOs, the order in which the
UPO’s are loaded is also important. The layout order determines which UPO’s take
precedence on the VGO. Since there is the possibility of loading two UPO’s in the
same space on the VGO, the layout order determines which UPO should be loaded
last (i.e., be the last to adjust the VGO). Once the UPO-based volumes are loaded on
the VGO for a particular frame, each ray can compute the atmospheric profile along
the ray.
can generate the atmospheric profile for any ray. IRIMAGE uses the stack of grid
point indexes stored by each ray and retrieves the gas and aerosol model database
indexes for each grid point. These indexes are placed in an array along with the path
length through the associated voxel. IRIMAGE processes this new array until it has
reduced the voxel stack to a smaller set of atmospheric layers which have different gas
and/or aerosol models. Each layer is the combination of several adjacent voxels that
share the same gas and aerosol index. IRIMAGE just combines these voxels into a
single layer whose path length is the sum of all the affected voxels. Once the profile is
generated, it is passed onto MODTRAN which computes the radiance along the ray’s
path. From this radiance, IRIMAGE is able to determine the output signal from a
metric models. The VGO was intially set to be a standard 1976 U.S. Atmosphere at
293K and 760 torr. Each UPO model used the same standard atmosphere parameters
as the default conditions loaded on the VGO except that the temperature was 350K
and the concentration of Carbon Monoxide (CO) was raised to 50,000 part per million
5% carbon monoxide at 350K in a standard 1976 U.S. atmosphere at 293K. (a) Box,
(b) Cone, (c) Cylinder, (d) Ellipsoid.
293K and had standard levels of CO. The levels of CO in a standard atmosphere are
between 0.05 to .2 ppm according to Seinfeld [11]. In addition, Seinfeld shows that
the heightened CO level we used for the bounded volumes is typical for the output of
an internal combustion engine prior to the exhaust entering the catalytic converter.
The background imaging VGO had a fixed temperature of 293K and an emissivity
of .9 (90% of a blackbody). The various images generated clearly demonstrate how
a bounded volume based on the UPO construct can be loaded onto the VGO. In
addition, one can see how the ray-VGO intersection algorithm creates a reasonable
Although major or minor modifications can be made to either object, these changes
won't effect how other routines use the VGO or the UPO as long as the modifica-
tions don’t effect how their routines are called. This reduces the amount of recoding
required to make a change or improvement. For instance, the VGO has no direct
relationship to the various objects (image, UPO, etc.) placed upon it. Therefore,
new geometric models and interpolation algorithms can be implemented which do
not affect the internal structure of the VGO. At the same time, different algorithms
for accessing the FDA or the OIA can be implemented which add new capabilities
to the VGO and do not affect its current routines. Besides modifying the functional-
ity of the VGO and the UPO, their object-oriented design provides the user and/or
programmer with the flexibility to define new primitive surface models or custom
forms of the VGO. A programmer can tailor certain aspects of the VGO based on the
requirements of the application, the speed of the machine and the memory available
for the program. The user may add new types of primitives to the UPO library by
developing ways to define these primitives using the UPO’s methods of connecting
contours. Another advantage to an object-oriented approach is the multiple reuse of
uses three different implementations of the VGO: a 2D background imaging VGO, a
3D Atmosphere VGO, and a 2D FPA layout VGO. Using a piece of code multiple
times not only cuts down on how much code is written, but also reduces the amount
of memory used by the program to store the various routines. Overall, these factors
make the VGO and the UPO good examples of useful tools which can be maintained
and improved without affecting their functionality.
ray tracing method is effective at determining an approximate profile, but the voxel
nature prevents the reflective and refractive advantages of a ray trace from being ex-
ploited. This limits the VGO’s usefulness to applications which do not require these
properties. In addition, the regular spacing of grid points can create unnatural effects
for complex and slow moving objects (jerkiness, “flashing pixels,” etc.) These prob-
lems are inherent with any representation of space which uses a voxel approximation.
However, most of these problems can be avoided or minimized with good program
design and careful planning of how to use the VGO.
quite well. It sufficiently flexible to handle any type of geometry that can be built
using the UPO as a basis. However, the algorithm still has room for improvement.
Currently, the scale and number of grid points determines the accuracy of the interpo-
lation of the bounded volume. Large numbers of grid points spaced close together give
excellent representations of the volumes but at the cost of huge amounts of memory.
Smaller numbers of grid points spaced far apart will severely limit the accuracy of the
model for most cases, but it limits the amount of memory used and reduces the time
required to compute the interpolation. Unfortunately, it is very hard to determine
the perfect balance. The accuracy is also limited by the fact that a single cutting
plane determines how an entire plane of grid points is interpolated. Multiple planes
could be implemented that would more accurately determine what perentage of the
volume of the voxel is contained inside the bounded volume instead of using a single
area value. Finally, the current interpolation algorithm does provide a method of
can attempt to approximate this property by loading “mixed” data sets and placing
them at these boundries.
a new approach to modeling three-dimensional scenes. Although there are certain
drawbacks, the object-oriented design and their more generalized nature make it
possible to improve their capabilities over time. The additional techniques of ray
tracing through the VGO, and the simulation of temporal behavior exploited by
IRIMAGE, demonstrates that this methodology is effective in simulating behavior
Computer Graphics. (Addison-Wesley, Reading, MA, 1990), pp. 377-81.
and J. Hughes. Computer Graphics: Priciples and Practice, Second Edition.
(Addison-Wesley, Reading, MA, 1990), pp. 557, 712-714.
Revisted,” Int. J. Numer. Methods in Engineering. 37, 1990, 3605-3619.
Computer Graphics. (Addison-Wesley, Reading, MA, 1990), pp. 382-83.
Computer Graphics. (Addison-Wesley, Reading, MA, 1990), pp. 180-182.
Sheffield, July 1990.
of Rectangular Grid Blocks within a Polygon,” Computers & Geosciences, 16(3)
1990, 379-384.
L.S. Bernstein, D.C. Robertson, P.K. Acharya, L.S. Rothman, J.E.A. Selby,
7 Model,” prepared for PL/GPOS, 1996.
& Sons, New York, 1986).
Used by IRIMAGE
IRIMAGE to run a simulation. In this case, these are the parameter files for the Car
simulation discussed in chapter 2.5. There are seven files, a single general file and
one for each of the major objects of IRIMAGE. You should notice that the format of
C.1 The General Parameter File
Memory Model = 0
Total Number of Frames= 90
Scene Parameter File = car_sim/script/car.scn
Atmosphere Param File = car_sim/script/car.atm
Detector Param File = car_sim/script/car.det
FPA Parameter File = car_sim/script/car.fpa
Optics Parameter File = car_sim/script/car.opt
Qutput Info to Screen = 0
Output Info to File =0
C.2 The Scene Object Parameter File
Scene Grid Size = 400, 400
Interpolation Method = 0
Background Mask = 0.0
Use Animation Param. = 1
First or Only Frame = i
Image Index = 1
Corner Location = 1
Grid Point:X,Y Sizing = 200, 200
Pos. of Emissivity
Pos. of Mask Values
Image File Name
Type of Image File
Image Index
Pos. of Emissivity
Pos. of Mask Values
Binary:Temper. Mult
Binary:Mask Add
Image File Name
Type of Image File
Image Index
Grid Point:X,Y Sizing
Pos. of Emissivity
Pos. of Mask Values
400, 300
car_mat.ascii
400, 300
Number of VIOs
VIO Object Name
VIO Object Index
VIO Object Index
Transparency at Min.
Maximum value of Mask
VIO Object Index
Car_Matte
Interpolation Method
Frame Number
Orient. about Origin
Scale about Origin
# of FP Data Pts
$ Key Frame Param: Car
Number of Key Frames
Interpolation Method
Frame Number
Orient. about Origin
Scale about Origin
# of FP Data Pts
Frame Number
Orient. about Origin
Scale about Origin
Interpolation Method
Frame Number
Object Active
Origin Location (WCS)
.0, 0.0, 180.0
0.0, 0.0, 0.0
1.0, 1.0, 0.0
0.0, 0.0, 0.0
0.0, 0.0, 0.0
# of Interp. Coeff. = 0
Frame Number = 90
Object Active =1
Orient. about Origin = 0.0, 0.0, 0.0
Scale about Origin = 1.0, 1.0, 0.0
Object Type Pos #0001
Object Type Pos #0001
Num. of Models in DB
Temperature Value/Unit=
H20 Density/Unit =
02 Density/Unit =
03 Density/Unit
.931e+02, A
NO Density/Unit =
N20 Density/Unit =
NH3 Density/Unit =
HNO3 Density/Unit
[> <> a > > <> OP)
rn DDD DH YD
Model Index: 2
Temperature Value/Unit= 3.232e+02, A
H20 Density/Unit = 0.0, 6
02 Density/Unit = 0.0, 6
03 Density/Unit = 0.0, 6
CO Density/Unit = 5.00e+03, A
CO2 Density/Unit = 0.0, 6
CH4 Density/Unit = 0.0, 6
NO Density/Unit = 0.0, 6
NO2 Density/Unit = 0.0, 6
N20 Density/Unit = 0.0, 6
NH3 Density/Unit = 0.0, 6
HNO3 Density/Unit = 0.0, 6
S02 Density/Unit = 0.0, 6
Rain Rate = 0.0
Aerosol Model = 0
Cloud Model Type = 0
Upper Atm Aerosol Mod.= 0
Default
Default
Default
Default
Default
Default
T/P Model
Trace Gas Mdl
0.
(> > oP a
CO2 Density/Unit =
03 Density/Unit =
N20 Density/Unit =
CO Density/Unit =
CH4 Density/Unit =
02 Density/Unit =
NO Density/Unit =
S02 Density/Unit =
NO2 Density/Unit =
NH3 Density/Unit =
HNO3 Density/Unit
Layer Temperature/Unit=
H20 Density/Unit =
C02 Density/Unit =
03 Density/Unit =
N20 Density/Unit =
CO Density/Unit =
CH4 Density/Unit =
02 Density/Unit =
NO Density/Unit =
S02 Density/Unit =
NO2 Density/Unit =
NH3 Density/Unit =
HNO3 Density/Unit =
-60e+02, C
-931e+02, A
-61e+02, C
-931e+02, A
oO
Layer Temperature/Unit=
H20 Density/Unit =
CO2 Density/Unit =
03 Density/Unit =
N20 Density/Unit =
CO Density/Unit =
CH4 Density/Unit =
02 Density/Unit =
NO Density/Unit =
S02 Density/Unit =
NO2 Density/Unit =
NH3 Density/Unit =
Layer Temperature/Unit=
H20 Density/Unit =
CO2 Density/Unit =
03 Density/Unit =
N20 Density/Unit =
CO Density/Unit =
CH4 Density/Unit =
02 Density/Unit =
NO Density/Unit =
S02 Density/Unit =
NO2 Density/Unit =
NH3 Density/Unit =
HNO3 Density/Unit =
Initial Altitude =
Final Altitude =
Initial Zenith Angle =
Path Length =
-931e+02, A
DDD Da DDH DDH DD DD
« . “ . . . ® .
NO
nD DDD DDH DH DH HH HD DW
Type of Path =1
Frequency Interval WN = 2
Fullwidth at Half Max = 2
Ellipse Type Index = 1
Cone Type Index = 2
Cylinder Type Index = 3.
Cone Model Index = 1
Height along Z axis = 0.4
Radius of Base = .075
X Offset of Top = 0.0
Y Offset of Top = 0.0
$ Geom Model: Cylinder DB
Num of Cylinder Models= 0
Interpolation Method = 0
Frame Number = 1
Object Active = 1
Orient. about Origin = 0.0, 90.0, 0.0
Scale about Origin = 0.75, 0.75, 1.0
Integer Data Value #02
# of FP Data Pts =
Frame Number =
Integer Data Value #02
# of FP Data Pts =
# Layout DB Loads Object
be No
0.0, 90.0, 0.0
1.0, 1.0, 1.0
Layout Key Frame Num. =
Num of Objects in Fram=
C.4 The FPA Object Parameter File
WCS Orientation of FPA= 0.0, 0.0, 0.0
Optics Object Index = 1
Output Object Index = 1
Unit Cell:# of Det(X,Y)= i, 1
Cell Loc. in Unit Cell= 0, 0
Detector ID Index =1
Bandwidth (wavenumber)= 2
C.5 The Detector Object Parameter File
Number of PC Detect = 0
Number of Pyro Detect = 0
Number of Bolo Detect = 0
Quantum Efficiency = 0.65
Load Resistance = 1.66e9
Integration Time = 0.005208
Detector Temperature = 77.0
Sigma Width
Location of curve
Output Signal Units
Input Power Units
Spectral Unit
Responsivity/D* Value
Spectral Unit
Responsivity/D* Value
Spectral Unit
Responsivity/D* Value
Spectral Unit
22.2
548.57
.000911
1.0e-20
1850.0
1.90e-7
2000.0
1.90e-7
3400.0
1.90e-7
# Simulation Name: Car
# Created on Date: 9/3/97
# Optics Parameters (optics.cxx)
# Load Specific Optical Index (id) Right Justify
$ Optical Param # 1
Front OA Vector (WCS) = 0.0, 0.0, -1.0
1st Focal Point(WCS) = 0.0, 0.0, 0.07
1st Principal Pnt(WCS)= 0.0, 0.0, -0.03
1st Nodal Point(WCS) = 0.0, 0.0, -0.03
Back Origin (WCS) = 0.0, 0.0, -0.1
Back OA Vector (WCS) = 0.0, 0.0, -1.0
2nd Focal Point(WCS) = 0.0, 0.0, -0.17
2nd Principal Pnt(WCS)= 0.0, 0.0, -0.07
2nd Nodal Point(WCS) = 0.0, 0.0, -0.07
Nodal Phase Angle = 0.0
Transmission Coeff Loc= 2
# of Transmission Pnts= 12
Frequency (wavenumber)= 1800
Transmission Coeff. = 0.935
Frequency (wavenumber)= 2000
Transmission Coeff. = 0.935
Frequency (wavenumber)= 2083
Transmission Coeff. = 0.930
Frequency (wavenumber)= 2174
Transmission Coeff. = 0.920
Frequency (wavenumber)= 2273
Transmission Coeff. = 0.920
Frequency (wavenumber)= 2381
Transmission Coeff. = 0.925
Frequency (wavenumber)= 2500
Transmission Coeff. = 0.930
Frequency (wavenumber)= 2632
Transmission Coeff. = 0.945
Frequency (wavenumber)= 2778
Transmission Coeff. = 0.930
Frequency (wavenumber)= 2941
Transmission Coeff. = 0.910
Frequency (wavenumber)= 3125
Transmission Coeff. = 0.910
Frequency (wavenumber)= 3333
Frequency (wavenumber)= 2000
Transmission Coeff. = 0.0
Frequency (wavenumber)= 2973
Transmission Coeff. = 0.0
Frequency (wavenumber)= 2993
Transmission Coeff. = 0.050
Frequency (wavenumber)= 2999
Transmission Coeff. = 0.100
Frequency (wavenumber)= 3005
Transmission Coeff. = 0.200
Frequency (wavenumber)= 3014
Transmission Coeff. = 0.700
Frequency (wavenumber)= 3028
Transmission Coeff. = 0.775
Frequency (wavenumber)= 3044
Transmission Coeff. = 0.730
Frequency (wavenumber)= 3053
Transmission Coeff. = 0.680
Frequency (wavenumber)= 3065
Transmission Coeff. = 0.710
Frequency (wavenumber)= 3067
Transmission Coeff. = 0.700
Frequency (wavenumber)= 3071
Transmission Coeff. = 0.600
Frequency (wavenumber)= 3082
Transmission Coeff. = 0.200
Frequency (wavenumber)= 3086
Transmission Coeff. = 0.100
Frequency (wavenumber)= 3091
Transmission Coeff. = 0.050
Frequency (wavenumber)= 3113
Transmission Coeff. = 0.0
Frequency (wavenumber)= 3333
Frequency (wavenumber)= 2000
Transmission Coeff. = 0.0
Frequency (wavenumber)= 2833
Transmission Coeff. = 0.0
Frequency (wavenumber)= 2861
Transmission Coeff. = 0.025
Frequency (wavenumber)= 2886
Transmission Coeff. = 0.050
Frequency (wavenumber)= 2899
Transmission Coeff. = 0.100
Frequency (wavenumber)= 2915
Transmission Coeff. = 0.850
Frequency (wavenumber)= 2928
Transmission Coeff. = 0.901
Frequency (wavenumber)= 2941
Transmission Coeff. = 0.893
Frequency (wavenumber)= 2967
Transmission Coeff. = 0.901
Frequency (wavenumber)= 3030
Transmission Coeff. = 0.885
Frequency (wavenumber)= 3110
Transmission Coeff. = 0.915
Frequency (wavenumber)= 3210
Transmission Coeff. = 0.890
Frequency (wavenumber)= 3231
Transmission Coeff. = 0.850
Frequency (wavenumber)= 3263
Transmission Coeff. = 0.200
Frequency (wavenumber)= 3279
0.050
3311
0.025
3344
0.0
4000
0.0
2000
0.0
2949
0.0
2965
0.050
2970
0.100
2974
0.200
2990
0.800
2999
0.862
3013
0.862
3018
0.825
3028
0.790
3033
Frequency (wavenumber)= 3053
Transmission Coeff. = 0.200
Frequency (wavenumber)= 3064
Transmission Coeff. = 0.050
Frequency (wavenumber)= 3085
Transmission Coeff. = 0.0
Transmission Coeff. = 0.0
Frequency (wavenumber)= 2880
Transmission Coeff. = 0.0
Frequency (wavenumber)= 2896
Transmission Coeff. = 0.050
Frequency (wavenumber)= 2903
Transmission Coeff. = 0.100
Frequency (wavenumber)= 2923
Transmission Coeff. = 0.750
Frequency (wavenumber)= 2929
Transmission Coeff. = 0.788
Transmission Coeff. = 0.700
Frequency (wavenumber)= 2972
Transmission Coeff. = 0.050
Frequency (wavenumber)= 2985
Transmission Coeff. = 0.0
$ Transmission Coeff # 6
Frequency (wavenumber)= 2670
Transmission Coeff. = 0.0
Transmission Coeff. = 0.025
Frequency (wavenumber)= 2682
Transmission Coeff. = 0.050
Frequency (wavenumber)= 2687
Transmission Coeff. = 0.100
Frequency (wavenumber)= 2692
Transmission Coeff. = 0.200
Frequency (wavenumber)= 2703
Transmission Coeff. = 0.665
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber) =
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber) =
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
# of Transmission Pnts=
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber )=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
2721
0.700
2726
0.600
2736
0.200
2741
0.100
2745
0.050
2752
0.025
2762
0.0
3333
0.0
2000
0.0
2032
0.0
2539
0.025
2545
0.050
2550
0.100
2556
0.200
Transmission Coeff. = 0.785
Frequency (wavenumber)= 2579
Transmission Coeff. = 0.796
Transmission Coeff. = 0.200
Frequency (wavenumber)= 2602
Frequency (wavenumber)= 2605
Frequency (wavenumber)= 2612
Transmission Coeff. = 0.025
Frequency (wavenumber)= 2620
Transmission Coeff. = 0.0
Frequency (wavenumber)= 2433
Transmission Coeff. = 0.025
Frequency (wavenumber)= 2451
Transmission Coeff. = 0.050
Frequency (wavenumber)= 2463
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber) =
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber) =
# of Transmission Pnts=
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber) =
Transmission Coeff. =
Frequency (wavenumber) =
2509
0.872
2522
0.898
2545
0.870
2567
0.885
2581
0.870
2594
0.800
2635
0.100
2653
0.050
2667
0.025
2740
0.0
3333
0.0
2000
0.0
2326
0.0
2335
0.025
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber) =
Transmission Coeff. =
Frequency (wavenumber) =
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber) =
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
$ Transmission Coeff #10
# of Transmission Pnts=
Frequency (wavenumber)=
Transmission Coeff. =
0.050
2345
0.100
2360
0.700
2369
0.750
2372
0.754
2375
0.750
2378
0.700
2392
0.100
2396
0.050
2404
0.020
2410
0.0
3333
0.0
2000
0.0
2283
0.0
2288
Transmission Coeff. =
Frequency (wavenumber) =
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber) =
Transmission Coeff. =
Frequency (wavenumber) =
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
2294
0.020
2296
0.040
2299
0.080
2304
0.200
2320
0.770
2326
0.840
2336
0.870
2342
0.860
2347
0.830
2353
0.810
2364
0.840
2370
0.850
2375
0.820
2387
0.670
2389
0.650
2392
0.660
2398
0.740
2401
Frequency (wavenumber) =
Transmission Coeff. =
Frequency (wavenumber) =
Transmission Coeff. =
Frequency (wavenumber) =
Transmission Coeff. =
Frequency (wavenumber) =
Transmission Coeff. =
Frequency (wavenumber) =
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber) =
$ Transmission Coeff #11
# of Transmission Pnts=
Frequency (wavenumber) =
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber) =
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber) =
2404
0.760
2415
0.250
2421
0.060
2427
0.025
2433
0.010
2439
0.004
2451
0.0
3333
0.0
2000
0.0
2193
0.0
2200
0.010
2212
0.025
2217
0.050
2221
0.100
Transmission Coeff. =
Frequency (wavenumber) =
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber) =
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber) =
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber) =
0.200
2228
0.300
2235
0.800
2237
0.830
2239
0.857
2247
0.830
2250
0.800
2260
0.300
2262
0.200
2263
0.100
2270
0.050
2275
0.025
2283
0.010
2294
0.0
3333
0.0
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber) =
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
2000
0.0
2128
0.0
2167
0.025
2176
0.050
2181
0.100
2195
0.595
2210
0.795
2220
0.815
2227
0.795
2250
0.595
2270
0.795
2275
0.595
2288
0.100
2291
0.050
2298
0.025
2326
0.0
3333
0.0
$ Transmission Coeff #13
# of Transmission Pnts=
Frequency (wavenumber )=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber) =
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber )=
Transmission Coeff. =
Frequency (wavenumber)=
Transmission Coeff. =
Frequency (wavenumber) =
Transmission Coeff. =
Frequency (wavenumber) =
Transmission Coeff. =
Frequency (wavenumber )=
1800
0.0
1951
0.0
1992
0.025
2010
0.100
2058
0.800
2073
0.850
2083
0.840
2141
0.875
2183
0.900
2227
0.9222
2270
0.900
2281
0.800
2370
0.100
2378
0.050
2392
0.025
Transmission Coeff.
Frequency (wavenumber )
Transmission Coeff.
Frequency (wavenumber)
Transmission Coeff.
Frequency (wavenumber)
Transmission Coeff.
Frequency (wavenumber)
Transmission Coeff.
Frequency (wavenumber)
Transmission Coeff.
Frequency (wavenumber)
Transmission Coeff.
Frequency (wavenumber)
Transmission Coeff.
Frequency (wavenumber)
Transmission Coeff.
Frequency (wavenumber)
Transmission Coeff.
Frequency (wavenumber)
Transmission Coeff.
Frequency (wavenumber)
Transmission Coeff.
Frequency (wavenumber)
Transmission Coeff.
0.0
3333
0.0
2000
0.0
2054
0.0
2059
0.010
2071
0.025
2076
0.050
2084
0.100
2089
0.200
2104
0.700
2107
0.750
2112
0.780
2116
0.770
2125
0.785
2133
Frequency (wavenumber)= 2139
Transmission Coeff. = 0.730
Frequency (wavenumber)= 2145
Transmission Coeff. = 0.720
Frequency (wavenumber)= 2156
Transmission Coeff. = 0.750
Frequency (wavenumber)= 2162
Transmission Coeff. = 0.725
Frequency (wavenumber)= 2177
Transmission Coeff. = 0.200
Frequency (wavenumbe:”)= 2182
Transmission Coeff. = 0.100
Frequency (wavenumber)= 2184
Transmission Coeff. = 0.050
Frequency (wavenumber)= 2191
Transmission Coeff. = 0.025
Frequency (wavenumber)= 2203
Transmission Coeff. = 0.0
# Filter - (Cen = , FWHM = )
# of Transmission Pnts=
Frequency (wavenumber)=
$ Output Parameters # 1
Output Filename Head =
Qutput Detector Type =