Keywords: lidar, jitter, motion
Summary
This demo focuses on the jitter and knowledge error tools available in the DIRSIG model in the context of a LIDAR system.
Details
Both passive and active systems are impacted by jitter or "unprogrammed motion" because it is a deviation from the desired or commanded platform location, platform orientation or platform-relative pointing. These types of deviations can be due to platform vibrations or environmental conditions (turbulence, wind, etc.). We typically call the deviation from the desired state (location, orientation, pointing, etc.) to this actual state to be a "command error".
In an attempt to measure command errors, we usually attach things like GPS and INS instruments to the platform to try to measure what actually happens. However, these instruments are not flawless and they have errors (quantization, noise, temporally course, etc.). We call the difference between the actual state and the measured state a "knowledge error".
Command errors are not desired because they mean we don’t image what we wanted to. However, as long as we don’t have knowledge errors, we will know where we imaged. Knowledge errors manifest themselves as thinking you imaged location X when you actually imaged location Y.
Command errors are incorporated into the DIRSIG simulation as platform location, platform orientation or platform-relative pointing jitter. Knowledge errors are incorporate into the final LIDAR products in the LIDAR Detector model.
Important Files
-
There are a pair of
.platformfiles. One without any jitter on the fixed (static) mount and one with 1 mrad of along-track and across-track jitter.-
The
without_jitter.platformandwithout_jitter.simfiles produce thewithout_jitter.binfile. -
The
with_jitter.platformandwith_jitter.simfiles produce thewith_jitter.binfile.
-
Setup
This simulation consists of a 32 x 32 array LIDAR mounted to a platform that is statically positioned (a single location and orientation entry) over the scene. The jitter causes the LIDAR to change platform-relative pointing for each pulse. The scene is composed of a flat plate with a set of 1 x 1 meters (various height) box targets on it. The boxes are arranged in a 3 x 3 cross pattern with a 3 meter tall box at the center, 2 meter tall boxes adjacent to the center box and 1 meter tall boxes at the edges.
Radiometric Simulation
The user should run the two simulations in the demo, which will produce a BIN file that reflects no jitter during the collection and a BIN file that had jitter during the collection.
Lidar Detection
The two BIN files were run through the LIDAR detector model to produce ASCII/Text point cloud files. The examples presented here were run with a Geiger mode detector with ideal detection (PDE = 1) and no noise (DCR = 0). However, a Linear mode system would reproduce the same effects. These scenarios specifically explore jitter and knowledge of that jitter which is largely independent of the detection method.
Each BIN file (one with jitter and one without) is run twice: once with knowledge errors and once without. This results in four 3D point clouds.
Results
Without Jitter and Prefect Knowledge
In this scenario, the platform without mount jitter is run and the LIDAR detector model is run without any knowledge errors. As a result, we get a clean rendering of the scene. The low point fill (low point density) is because all 25 pulses image the exact same locations. Hence, each point is really 25 points (one from each pulse) directly on top of each other.
Without Jitter and Knowledge Errors
In this scenario, the platform without mount jitter is run and the LIDAR detector model is run with 0.1 mrad of along-track and across-track platform-relative pointing (mount) knowledge errors. As a result, we get a blurring of the box targets.
With Jitter and Prefect Knowledge
In this scenario, the platform with 1 mrad of mount jitter is run and the LIDAR detector model is run without any knowledge errors. Because of the pointing jitter, we no longer image the exact same locations 25 times (like in the first scenario). The shape and location of the box targets is preserved (crisp edges) because the LIDAR detector model has precise knowledge of the jitter during the collection. As a result, we get a clean rendering of the scene with increased fill (higher point density) compared to the first scenario. Note that the overall footprint of the scene is slightly larger.
With Jitter and Knowledge Errors
In this scenario, the platform with 1 mrad of mount jitter is run and the LIDAR detector model is run with 0.1 mrad of knowledge errors. Like the second scenario, the box targets get blurred by the errors in pointing knowledge.
