Fast Segmentation of 3-D Point Clouds Based on Ground Plane State Tracking

3D point cloud segmentation is the first and essential step for LIDAR-based perception, and its result has a great impact on subsequent tasks such as classification and tracking. This paper proposes a fast and precise two-step 3D point cloud segmentation algorithm based on ground plane state tracking. The algorithm first extracts the points belonging to the ground. To avoid processing the whole point cloud in every frame and improve computational efficiency of the algorithm, we introduce tracking idea into the ground point extraction step and estimate the ground plane state by the fusion of prior information and measurement information. the second step is to cluster the remaining non-ground points. We introduce an adaptive threshold-based RBNN (Radially Bounded Nearest Neighbor strategy) clustering algorithm which reduces the number of mis-segmentation by determining the adaptive threshold function according to the characteristics of Lidar point cloud. Experimental results demonstrate that the algorithm consumes less time and achieves high segmentation accuracy than previous works.