Multi-Threshold based Ground Detection for Point Cloud Scene

Point cloud is widely used in self-driving technology recently. Usually the first step of point cloud processing is segmentation of ground points and non-ground points. In this paper, a multi-threshold detector is proposed for point cloud scene captured by LiDAR mounted on an autonomous vehicle. The proposed algorithm uses variant thresholds which depend on the distance between two consecutive points. Furthermore, the algorithm also proposes additional rules for finding the start ground point of each scanning line and eliminating the backward slope. Simulation result shows the proposed algorithm works well in different testing environments, in terms of miss rate, accuracy and execution time. For one scene with more than 180,000 points, the segmentation can be done in 8 ms and with 99.5% accuracy rate by the proposed algorithm.