A Geometric Feature Based Traversable Area Extraction and Evaluation Method for 3D Point Clouds

Abstract

This paper aims to provide a light and practical point cloud map for ground robots. A novel geometric feature based method is proposed. First, the 3D point cloud map is pre-processed, then the traversable area is extracted by a fusion method consisting of model fitting, filtering, and nearest neighbor search. Terrain flatness and boundary risk index are proposed to evaluate the traversable status of different terrain points. Based on the results of the extraction of traversable area, the geometric location and the neighbor point clouds are used to perform terrain assessment. The experimental results show that our approach can extract the traversable area from the original 3D point cloud map, with the map size and the number of points within the map greatly reduced, and the terrain points can be distinguished by traversability values.