DIRSIG: A Framework for Radiometry Modeling and Image Simulation

1
DIRSIGA Framework for Radiometry Modelingand
Image Simulation
2
Overview

• What is it?
• DIRSIG is a toolbox of first principles
  radiation propagation models with a focus on the
  prediction of photon travel through scenes at
  spectroscopic resolutions.
• In most cases, that toolbox is used to model an
  imaging system an generate images.
• What do we do with it?
• The model is used to predict the radiance from
  arbitrary paths or targets in a complex
  environment.
• These predictions can be driven by a sensor model
  (built-in) to form images.
• Why do we do it?
• Exploring the physics required to accurately
  predict real world observations provides us with
  insights to the image formation process as a
  whole.
• This knowledge and understanding of phenomenology
  can be used to make better models and to make
  smarter algorithms.
• Availability
• The DIRSIG model is available as a binary
  distribution for most UNIX-based platforms for
  any U.S. government organization or contractor.

3
Object Geometry
Themodynamic Optical Properties
Ray Tracer
Plume Description
Thermal Model
Weather Database
Plume Model
DIRSIG Executive
Optical Property Models
Radiometry Model
Focal Plane Description
Atmospheric Model
Sensor Model
Platform Description
MODTRAN MODTRAN-P
FASCODE
4
Applicable Areas

• Sensor Prototyping
• Construct and evaluate sensor designs in a
  virtual environment
• Produce example products for customers
• Algorithm Testing
• Create custom data sets to test algorithms
• Control all the image formation variables and use
  per-pixel truth to better evaluate performance.
• Algorithm Training
• Use the model to construct inputs to
  physics-based algorithms.
• Predict the radiance of a target in various
  conditions.
• Analyst Training
• Create custom training examples to test the
  sensitivity of the analyst to phenomenology
• Provide a hypothesis tool to explore the nature
  of phenomenology

5
Atmospheric Database
Thermodynamic Optical Properties
Sensor Model
Weather Database
Object Geometry
Thermal Model
Radiometry Model
DIRSIG
Target and background databases with spatial and
spectral variability (clutter)
Fully-spectral radiation propagation
Polarized/Unpolarized Broadband, multi-, hyper-
or ultra-spectral imagery
Simulated Data Products
Thermal IR and Low-Light Imagery
Hyperspectral Imagery and Target Maps
Range Gated LADAR
Simulated Data Exploitation Products
Obscured Target Detection
Image Fusion Demonstrations
Sub-Pixel Detection Performance (ROC curves)
6
MegaScene1 Rochester, NY
IKONOS Color Image

• 25,000 objects
• 5.5 billion facets
• 8.0 km2 (3.0 mi2)

DIRSIG Simulation
7
MegaScene 1
A channel from a thermal infrared hyperspectral
simulation featuring strong and weak gas plumes.
8
Data Fusion Prototyping
Image Fusion of Simulated Thermal and Low-Light
Imagery
9
Topographical LIDAR
DIRSIGPassive Imagery
Overhead
Slant View
DIRSIG DerivedTopo-Product
Slant View
Overhead
10
Framework Summary

• DIRSIG is an object-oriented infrastructure that
  provides a framework to model a collection of
  fundamental radiation propagation components that
  model the primary elements in the image chain
• Transmissive and non-transmissive surfaces and
  volumes
• Atmosphere, natural surfaces, man-made surfaces,
  plumes, leaves, camouflage nets, etc.
• The model framework supports several different
  propagation schemes to model multiple bounces
  and/or multiple scattering
• Backward ray-tracing with hierarchical ray
  sampling
• Forward and Backward coupled ray-tracing using
  photon mapping
• Full Backward Monte Carlo ray-tracing (unfunded
  development)

11
The Basic Framework
12
Scene Modules

• Traditional facetized geometry
• Produced with CAD and other programs and
  attributed for DIRSIG.
• Support for AutoCAD, Rhino3D, Blender and ONYX
  Tree.
• Support for models from off-the-shelf vendors
• Functional (non-facetized) geometry
• Plume models
• EPA/JPL Gaussian plume
• EPA CALPUF puff-based plume
• RIT gas cloud model
• RIT puff-based plume extended from Blackadars
  LaGrangian particle model
• Generic plume integration via voxelized interface
  
• Solid Geometry
• Boxes, spheres, etc.
• General geometry
• Height maps and bump maps
• Water surface models

13
Radiometry Modules (1)

• Temperature Prediction
• Integrated AIRSIM THERM
• Support to import PRISM/MuSES predictions
• Surface Reflectance Models
• Classic DIRSIG Diffuse/Specular (spatial-spectral
  variability (texture))
• Torrance-Sparrow BRDF
• Fresnel-based reflectance
• Beard-Maxwell BRDF and Sandford-Robertson BRDF
  (untested)
• Priest-Germer BRDF (Torrance-Sparrow
  polarization)
• Image-driven fractional mixing of an arbitrary
  number of materials
• Surface Transmission Models
• Fresnel-based reflectance
• Spectral transmission curves
• Surface Emissivity Models
• Spectral directional hemispherical emissivity
  (DHE) curves

14
Radiometry Modules (2)

• Bulk Absorption/Extinction Models
• Spectral absorption or extinction curves
• Bulk Emissivity Models
• Extinction derived emissivity
• Bulk Scattering Models
• Henyey-Greenstein parametric phase-functions
• Tabulated phase functions
• Atmosphere Models
• Simple, Mie-scattering based, parameterized
  atmosphere
• Gregg-Carder maritime atmosphere
• Uniform HydroLight (Mobley) maritime atmosphere
• MODTRAN- and/or FASCODE-based atmosphere

15
Hardware Modules

• Instrument Models
• Geometry Models
• Framing systems, pushbroom scanners, line
  scanners and whiskbroom scanners.
• Focal Plane and Read-Out Models
• Spectral response functions for pan, multi and
  hyperspectral systems.
• Filter based or dispersive (prism/grating) based
• Interference-based (Fourier Transform
  Spectrometer) models
• Michelson (temporal) and Sagnac (spatial) FTS
  models
• LADAR/LIDAR system models
• Full time-gated photon counts for user-defined
  listening windows.
• Instrument Pointing/Scaning Models
• Static, line-scan, whisk-scan and conical-scan
• User-defined or tabulated scan
• Platform Models
• Static and Dynamic platforms
• Position and orientation as a function of time
• Support for jitter and GPS/INS reporting

16
Radiometry Solvers

• Radiometry Solver objects are the algorithms that
  implement the computation of surface/volume
  leaving radiances.
• Different solvers implement different strategies
  to predict these results.
• One or more radiometry solvers are associated
  with every element in the scene.
• Depending on the simulation, multiple solutions
  may be desired
• The passive, environmentally-loaded radiance
• The active, reflected laser radiance (for LIDAR
  systems)
• The radiometry solver objects interact with the
  optical properties of the surface/volume through
  an abstracted interface.
• For example, the algorithm doesnt know if a
  surface has a reflectance model, or and
  emissivity model or both.

17
The Generic Radiometry Solver

• Used with backward, hierarchical ray-tracing
  method
• This solver predicts the surface radiance by
  sampling the hemisphere above the surface.
• That sampling is driven by the directional
  reflectance factors for the specific viewing
  geometry.
• If a surface has lobes in the BRDF, then these
  lobes are sampled more than the other parts of
  the hemisphere (e.g. importance sampling).
• The source-view geometry dependant reflectance
  (BRDF) is applied to each sample.
• Samples to important sources (Sun, Moon, etc.)
  are fixed but the user has control over the
  number of samples used in the rest of the
  hemisphere.
• The rays cast from this surface to sample the
  background will run the solvers for those
  intersected surfaces.
• The amount of the sampling on subsequent
  generations can be controlled by the user (hence,
  the hierarchical nature).
• The solution can be limited to a maximum number
  of bounces.

18
Background Interactions

• This solver focuses on illumination interactions
  between surfaces in the scene.
• Handling background loads in the surface
  radiative transfer reproduces interesting
  phenomenology including tree shine, sky
  shine, etc.

19
Applications and Phenomenology

• Loading Problems
• Sky, cloud and tree shine problems
• Cavity/Calibration Problems
• Cavity radiance effects from chambers/cavities
  with non-ideal surface properties.
• Polarization
• Full-spectral polarimetric modeling capability
• Currently heavily limited by available material
  characterizations
• Water/Littoral
• In-water scattering/absorption to understand
  adjacent effects.
• Plume Detection
• Provide a framework to evaluate instrument
  designs.
• Provide a source of data for rigorously testing
  algorithms.
• Truth is known for every pixel.

20
The MCDT Cavity Problem

• The radiance exiting the tower cowling is a
  result of the direct emission of the viewed
  surface and reflected emission from adjacent
  objects.
• To evaluate the expected variations in the tower
  leaving radiance, a model of the tower geometry
  can be constructed.
• The individual surfaces within the tower can be
  attributed with unique emissivity properties.
• DIRSIG can provide insight into the effect of
  uncertainties on the tower leaving radiance.
• Uncertainty in internal construction.
• Uncertainty in surface temperatures.
• Uncertainty in surface emissivity.

21
The Benefits ofFirst Principles Based Modeling

• Simulations demonstrating the power of DIRSIG
  coupled with MODTRAN.
• Single-scattering box of rain (left) and clouds
  (right) against a MODTRAN predicted sky.