Tri-Tzone: Tri-Grid Traversable Zone Segmentation based on Graph Searching Method using 3D LiDAR

Abstract

Traversable area for a ground vehicle includes terrains that the vehicle can explore, such as paved or unpaved roads, sidewalks, lawns, etc. In order to strictly determine the traversable area, slope, height, and continuity of terrain must be considered. In this paper, we introduce Tri-Tzone which enables traversable region segmentation on the 3D point cloud in real-time. A point cloud from a LiDAR sensor is encoded into TGF, which is a group of Tri-nodes. Then, the traversable nodes among the terrain nodes can be grouped by B-TGS. A final traversable terrain model is determined through the grouped nodes. We compare the proposed method with the state-of-the-art ground segmentation methods on CARLA simulator, which includes semantic labels for each points, and on our own challenging rough terrain data.