Transcript Brown_DIRSIG_Modeling

LADAR/LIDAR System Modeling
Scott D. Brown
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #1
Active Laser Sensing
Capability
y
x
Temporal Pulse
1.2
Power
1
0.8
0.6
0.4
0.2
0
4
3.5
3
2.5
2
1.5
1
0.5
0
Seconds
• The DIRSIG model has been enhanced to provide the user community
with a very flexible, active laser sensing capability.
– The synthetic world is stimulated with a pulsed beam that has a spatial,
spectral and temporal shape.
• Elaborate radiative transfer mechanisms reproduce important phenomenology
and “noise” sources in the simulated returns.
– The instrument model captures a time gated, photon arrival stream
• Dynamic instrument position and pointing can be incorporated.
• Spatially, spectrally and temporally oversampled data products can be produced
to drive back-end sensor models.
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #2
Active Channel Justification
• Historical focus on passive multi- and hyper-spectral systems
• Driving motivation to simulate active and passive sensors using the
same scenes and scenarios
– Perform trade studies between passive and active approaches
– Exploration of active/passive data fusion techniques
– Investigation of advanced exploitation algorithms
• Developed a challenging requirements set:
•
•
•
•
•
Rigorous atmospheric interactions
Participating mediums
Multiple bounce/scattering
Inclusion of passive returns
Complex scene geometries
• Moving platform and scanning effects
• Detailed optical descriptions
(BRDF and Scattering models)
• Arbitrary time-gated returns
• Mono & Bistatic configurations
• Prototype completed in 2002 by Burton and Brown
• Adapted prototype model and completed integration in 2004
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #3
End-to-End System Modeling
DIRSIG
Typical Products
Radiometry
Prediction
Geiger-Mode Systems
Focal Plane Reaching
Photon Counts
(as a function of time)
RIT GmAPD
Model
ITT GmAPD
Model
Single range
measurement
Linear-Mode Systems
No Detector Model
(yet)
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
First, last and other
range measurement
plus intensity
Slide #4
Research Affiliations
FastMetrix, Inc.
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #5
System Parameters
Laser/Source
Spectral Peak
Spectral Width
Beam Radius
Beam Divergence Angle
Beam Shape
Pulse Power
Pulse Duration
Pulse Period
Desired Photon Count
Beam Spread (On/Off)
Beam Wander (On/Off)
Image Wander (On/Off)
Focal Plane
Size in pixels
Oversampling
Spectral Region
Spectral Resolution
Response
07-Feb-2008
Instrument Mount
Mount Type
Scan Rate
Start Time Offset
Start Angle
Stop Angle
Instrument
Focal Length
Receiver Radius
Receiver Divergence Angle
Scan Rate
Signal Gate (Transmit, Receive, Wait)
RIT DIRSIG LADAR/LIDAR Modeling
Slide #6
Noise Sources
•
Passive/Environmental flux
–
The model continues to compute the passive (temporally constant) return from the
Solar/Lunar sources.
•
•
Photon arrival statistics
–
•
This includes atmospherically scattered (aerosol) photons.
Incorporate the appropriate uncertainty of photon arrivals in low-count situations.
Multiply-bounced source photons
–
Photons may arrive at the sensor at times that are correlated with longer ranges.
•
Possible problems when imaging clouds, plumes, tree canopies, etc.
6 ns
07-Feb-2008
7 ns
7 ns
RIT DIRSIG LADAR/LIDAR Modeling
Slide #7
Photon Starvation (1)
• Geiger-mode Avalanche Photo-Diode (GmAPD) LADAR
systems can operate in photon starved conditions.
– As a result, photon arrival and detection events are very rare.
• The detector technologies used in these cases cannot
guarantee the ability to detect every possible return event.
– For example, the APDs developed at MIT/LL have been shown to
follow Poisson detection statistics.
• But, this Poisson detection characteristic has benefits
– It allows for foliage penetration.
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #8
Photon Counts
Photon Starvation (2)
Leaf Return
Ground Return
• Consider the case of a single leaf over the ground.
– The leaf is reflective and transmissive (especially in the NIR region).
– The temporal return profile has two peaks
• One for the leaf return and one for the transmitted ground return.
– The detector might not trigger on the first peak, and trigger on the second
peak even though it is smaller.
• However, you need to shoot a lot of pulses to observe this low probability event.
• To achieve this you interrogate the same scene voxel numerous times.
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #9
Demonstrations
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #10
Tree Crown Demonstration
Scene Setup
Tree
Canopy
“Late”
Returns
Ground
Floor
Surface
Photon Map
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #11
Camouflaged Vehicle
Simulation
• Contains Material and Bump
Map made by thresholding
image of camouflage
– BM smoothed to create
gradients
• Material map contains a
“null” material (holes)
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #12
Camouflaged Vehicle
Simulation
Overhead Simulation
Digital Photo
Passive Simulation
DIRSIG LADAR Photon Map
High-fidelity 3D modeling and
photon mapping accurately
reveals camo net and vehicle
underneath for both active
and passive simulations
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #13
Camouflaged Vehicle
Simulation
Height Truth
Imagery
DIRSIG
CAD Models
LADAR
Pulse Cubes
Top of net
t003
Spreader
t012
t022
Humvee roof
t033
Humvee hood
t049
Humvee shadow
t115
Time
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #14
End-to-End System Simulation
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #15
End-to-End Topographical LADAR
Demo
• Microscene Data Collection (Mar
2004)
– MIT Lincoln Labs ALIRT System
•
•
•
•
•
Topographical LADAR
Sinusoidal whisk broom scanner
Scan FOV: 15 deg
Laser Peak  = 780 nm
Geiger-mode across-track scanned array
– Pixel Count: 32 x 32
– Nominal Flight Altitude ~1200 m
– Diverse Scene Content
Collection Scenario
• Trees, man-made objects, flat areas, etc.
• Some ground truth available
– Data basis for on-going validation effort
• Simulated real data collection for demo
purposes at this time
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #16
Modeling Process
Detailed Detector/Sensor Model
Time-Gated Arriving Photon
Counts Data Cube(s)
Fill Factor, Probability of Detection
Curves, Detector Response Curves,
MTFs, Dark Current, etc…
Topographic Products
Simulated Raw
Instrument Data
07-Feb-2008
Processing
RIT DIRSIG LADAR/LIDAR Modeling
Slide #17
LADAR at MicroScene1
Humvee
Grass on hill
Time/Distance
Shed Roof
07-Feb-2008
Portable
Generator
Shed
“Shadow”
RIT DIRSIG LADAR/LIDAR Modeling
“Late” photons that got “lost” in
grass
Lighter colored
dirt
Slide #18
DIRSIG
Passive Imagery
MicroScene1 LADAR Demo
Overhead
Slant View
Derived
Topo-Product
Topographic Products Courtesy of
Overhead
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slant View
Slide #19
“Spotlight” Collection Study
Concealed Targets - 45 deg Dwell
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #20
Tank with 60 Pulses
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #21
Tank with 240 Pulses
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #22
Tank with 480 Pulses
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #23
Geolocation Errors
• A set simulations using the DIRSIG instrument mount and
platform uncertainty features to demonstrate the impact of
uncertainty on geolocation and interpretibility.
• Scene consists of a 3D bar target with “tables” of different
sizes.
– The actual target exists at the MIT/LL flight facility and is routinely
collected by the MIT/LL ALIRT sensor.
~10 m
~26 m
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #24
Altitude Errors
• Comparison of baseline simulation and one with a 0.05
meter altitude error.
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #25
Cross-Track Pointing Errors
• Comparison of baseline simulation and one with a 0.5
mrad across-track pointing error.
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #26
Passive Illumination Effects
No Moon (11 PM)
Full Moon (4 AM)
• DIRSIG’s LADAR photon flux includes all the passive
radiometry of a “traditional” DIRSIG simulation.
– Passive illumination from Sun, Moon, sources can affect LADAR
performance.
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #27
Future Work
• Support for geo-located scenes and platforms (in progress)
• Track multiple-pulses in flight (in progress)
– Perform studies on the impacts of clouds
• Improved detector modeling
– More GmAPD effects (pixel crosstalk, arming triggers, etc.)
– Linear mode detector model
• Post-processing system trades
– Pulse-to-pulse total power variations
– Pulse-to-pulse temporal shape variations (e.g. multiple peaks)
• Improved user tools
– Data collection design tools (flight lines, collection scans, etc.)
– Sensor configuration and quick-look simulations.
• Enable and verify polarization
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #28
Summary
• DIRSIG has an active laser radar capability that has been
demonstrated for topographical LADAR and atmospheric
LIDAR problems.
– The model can be used in a variety of system engineering or asset
utility workflos.
• DIRSIG is a tool available to the government community
– Internally supported by RIT, no long term support contract.
– Quarterly software updates/releases
– Training courses
• DIRSIG development leverages commercial and
government funding.
– Everyone has access to same version of the model
– We team with anyone willing to contribute to the overall capability.
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #29
Extra Slides
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #30
Implementation
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #31
Possible Approaches
• Many models simply render a pulse return at the time or
range corresponding to the “target”.
– This doesn’t account for several effects:
•
•
•
•
Illumination obscurations (shadows).
Tilted surfaces that temporally stretch and skew returns.
Optical transmission through surfaces (e.g. glass, leaves, etc.)
Multiple bounces.
• DIRSIG does have an “preview” mode that will use this
approach to quickly give the user notional data.
– However, a more robust solution was needed for a rigorous
prediction and absolute radiometry.
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #32
Photon Mapping Pass #1:
Forward Photon Tracing
•
Photon “bundles” are forward, Monte-Carlo ray-traced from the source(s) into
the scene.
– A “bundle” is a set of photons that travel together.
– Russian Roulette techniques are used to determine absorption, reflected directions,
etc. based on material optical properties.
– Travel time is tracked throughout the photon’s flight.
•
Each event is recorded into the 3D “photon map” data structure.
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #33
Pass #1 Details
•
Bundles are shot from the source using a stochastic process
– Random position within the exit aperture
• Position density is constrained by user-defined spatial beam shape.
• Currently using direct method for Uniform/Flat/Tophat and Gaussian beams.
• Rejection sampling could be used for TEM profiles.
– Bundle direction is based on position and source divergence.
• Methods for turbulence driven divergence have been implemented.
– The beam centroid can be redirected in response to turbulence (e.g. beam wander)
– The source is mounted to an agile instrument mount (e.g. scanner) which is
mounted on an agile platform.
• Scanner/Mount positioning and platform noise can be modeled.
•
Other source properties
– Spectral shape is parametrically modeled (Gaussian and Lorentzian)
– Temporal shape is currently parametrically modeled.
• A tabulated temporal shape could be added.
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #34
Source Lasers
• Transverse Electro-magnetic (TEM) profile defines beam crosssectional intensity
– TEM 00 ≡ Gaussian beam profile (typically considered ideal)
Example laser TEM profiles
O’Shea, Donald C. Introduction to Lasers and Their Applications. Addison-Wesley, 1978.
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #35
Beam Density
•
Locations of ground arriving photon bundles from a uniform (left) and Gaussian
(right) beam densities with the same beam width.
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #36
Participating Mediums
Forward Scattering
Isotropic Scattering
A 2D side-looking view of the photon map with a box containing a
scattering medium hovering over the ground plane.
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #37
Photon Mapping Pass #2:
Photon Collection/Rendering
r
Surface Collection
Volume Collection
•
•
Rays are shot from focal plane plane and intersect surfaces
Photons are collected at each intersected point
– Search area/volume is determined by projected detector at range.
•
Individual bundles are redirected toward sensor based upon local optical
properties
– BRDF or scattering phase function depends on material type.
– Received bundles distributed and quantized for listening window and temporal
sampling frequency
•
Passive returns added by traditional radiance solvers
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #38
Pass #2: Collection
Sum,
Range-gate,
Apply
Phase
Function &
Identify
Events
Trace
Ray
Project
FOV
Recorded PhotonScattering
Map& Sample
Probability
from Detector
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #39
Pass #2 Details
• The range dependent projected area of the detector is used in the
collection search.
– That includes within a participating medium
– A surface search area is computed as a projected volume
• Geometry specific reflectance or scattering coefficients are used for
each photon bundle collected from the map.
– BRDF or scattering is potentially different for each bundle.
– Each bundle is absorbed within a medium based on its specific path length.
• Atmospheric returns analytically evaluated for efficiency
– An empirical solution is usually used as for the atmospheric returns.
• Direct calculation is very costly due to low probability of scattering/absorption
events in most atmospheres.
• The photon mapping approach can be used instead if the user desires.
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #40
Atmospheric Returns
• The scattering coefficient for a dry atmosphere might be
approximately 1x10-5 [1/m]
– Need to shoot 105 photons into a 1-meter long box to witness a
single scattering event.
• What is a realistic modeling scenario?
– Attempt to resolve vertical resolutions of a fraction of a meter.
– Interested in a path length of several thousand meters.
• How does that affect this approach?
– Need to model ~1010 photons within each spatial detector element
in order to witness a single scattering event within each numerical
contribution element.
– Ideally, a few orders of magnitude more to be statistically robust.
• What does that mean?
– Need an analytical solution
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #41
Atmospheric Returns
 R

c L A0
P( L ,R)  PL
( L )(R)( L ,R)exp2  ( L ,r)dr
2
2 R
 0

• Utilize the analytical solution proposed by Measures.
– For the ALIRT system, the returns are a fraction to a handful of a
photons integrated over the full 1,000 meter path.
• However, this depends on the type of atmosphere, altitude, etc.
– For different geometries (beam vs. detector FOV) or different
altitudes (path lengths), this component of the overall photon count
will be larger.
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #42
Atmospheric Properties
• Where do the atmospheric extinction and backscatter
coefficients come from?
– MODTRAN and/or FASCODE
– Only MODTRAN models scattering.
– MODTRAN5’s 1/10th wavenumber resolution is nearly small enough
for laser line transmission modeling.
• Backscatter coefficients are not normally output by
MODTRAN.
– Customized versions of MODTRAN are problematic.
– The DIRSIG4 make_adb tool computes the backscatter coefficients
as a function of altitude using a smaller differential volume
approach and a pair of MODTRAN runs.
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #43
Atmospheric Properties:
Extraction Methodology
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #44
Temporal Pulse Spread
Collection Area
Collection Area
On a tilted surface,
bundles have a range of
travel times
On a flat surface, all bundles
have the same travel time
Integrated
return
3 bundles on top of
each other
•
Integrated
return
3 bundles temporally
spread out
Important notes
–
Smoothness of the temporal shape for a tilted surface is dependant on the number
of bundles available in the collection area.
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #45
Temporal Pulse Spread
Scene Setup
10 deg  ~1m difference in range
Pulse width ~13.33 ns instead of 1.5 ns
DIRSIG accurately simulates the temporal pulse spreading
by linear convolution with impulse response of a scene
while still accommodating very high sampling rates
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #46
Other Sources of
Geolocation Errors
• Atmospheric turbulence along longer paths with deflect
light from the straight line path.
– Deflect the beam arriving at the ground from pulse to pulse.
• Sometimes referred to as “beam wander”.
• Not such a big deal as long as the projected beam overfills the
detector.
– Deflect the arrivals onto the focal plane.
• Sometimes referred to as “image wander”.
• The geolocation algorithms assume photons take straight paths.
• A deflected arrival will be recorded within the “wrong” pixel and
therefore will be incorrectly located within the scene.
• DIRSIG has some tools for modeling turbulence effects
including beam and image wander using Cn2
characterizations of the turbulence.
– Currently not available via the DIRSIG inputs files.
07-Feb-2008
RIT DIRSIG LADAR/LIDAR Modeling
Slide #47