Session: 06-07-01- Sustainable design

Paper Number: 96259

96259 - Assessment of LIDAR Imaging Systems for Autonomous Vehicles in Adverse Weather and Under Dynamic Conditions 

One of the main challenges for the safety validation of autonomous driving vehicles lies in the influence of weather phenomena. As each of the main sensors, namely LIDAR, radar, and cameras increases its sensitivity in order to detect smaller objects faster and hence be able to drive autonomously at higher speeds, the possible influence of environmental perturbations on their perception increases. Those perturbations could cause false positives, confuse self-calibration algorithms and reduce the sensor range. On the other hand, they could constitute a source of valuable information if the dependencies are known and properly characterized to better evaluate and predict road conditions or adapt its operation mode. The main purpose of this work is to evaluate the impact of adverse weather conditions (rain, fog, and snow) on LiDAR sensor performance in dynamic conditions, i.e. moving sensors and moving-target scenarios.

Two Lidars sensors, Velodyne VLP-32 and Cepton X90S were used in our investigation. In previous work [1] Velodyne VLP-32 and Ouster OS1-32 were used to investigate the effects of weather conditions on Lidar's performance in static conditions. Solid-State LiDAR (X90S) is used because it has the capability to overcome many limitations of spinning 3D LiDARs, and has irregular scan patterns.

 

Both Velodyne VLP-32 and Cepton X90S units communicate over an ethernet connection which makes it easy to interface with them. We wrote a program to analyze the data for moving platforms from our dynamic tests. The program includes commands for moving platforms in which sensors and targets are moving This program allows us to count the number of points in a region of space. We inspected each region before taking a reading to ensure that only returns from the target were counted.

 

To determine the impact of adverse weather conditions on the Lidar's performance, we recorded data in the rain, heavy snow, and fog. We analyzed our results using the effective range was determined. The object-detection algorithm is modified to allow a minimum number of points to detect an object reliably. We examined the effective ranges for a variety of thresholds under different weather conditions.

In this work, we are reporting the first set of data from the Velodyne lidar. More data will be presented later. Three tests have been done on a heavy snow day. The tests were done in a parking lot at Frostburg State University, Frostburg, MD, so the test can be easily recreated in snow, rain, and fog weather conditions using the parking spaces.

The first test had a car starting at 180 feet away from the LIDAR set-up and the car slowly moved about 5 mph until the car stopped about 10 feet from the LIDARs simply used to test the system. The second test had the car at the same position 180 feet away from the LIDAR system. The car move towards the LIDARs by increments of 18 feet (two parking spaces) until the car was directly in front of the LIDAR system. The third test was the same as test two except every time the car stops a person/pedestrian walked in front of the car at various distances. The purpose of the third test was to gain data that will be used later in the implementation of object tracking with objects in front of the vehicle being tracked.

The snow test results show a reduction in the recorded points. At 20m, 30% fewer returns were detected compared with a clear sunny day. At 50 m, the recorded points decreased by 35%. At 65 m and above, the recorded points were close to zero.

The pedestrian tests showed the same trend, however, the recorded points were 25% less compared to the car tests.

 

1.     J. Abdo, Spencer Hamblin, G. Chen (Published online Sep. 24, 2021), “Effective Range Assessment of Lidar Imaging Systems for Autonomous Vehicles Under Adverse Weather Conditions With Stationary Vehicles,” ASME J. Risk Uncertainty Part B ., Sep 2022, 8(3): https://doi.org/10.1115/1.4052228

 

Presenting Author: Jamil Abdo Frostburg State University

Presenting Author Biography: will be sent later

Authors:

Jamil Abdo Frostburg State University
Luke Russell University of Maryland-College Park
James Mills University of maryland - College Park
Taylor Frailey Frostburg State University
Genshe Chen Intelligent Fusion Technology, Inc.