A real-time grid map generation and object classification for ground-based 3D LIDAR data using image analysis techniques

A grid map generated from ground-based 3D LIDAR point clouds is a critical component for facilitating autonomous system navigation without crashing into obstacles and also to generate a road map for a large local area. This paper proposes a novel approach to generate an occupancy grid map along with object classification in real-time for autonomous ground robot applications. Based on geometric analysis of the raster-scanned LIDAR data, we formulate criteria to distinguish 3D points directly from coordinate values and generate three grid maps; occupancy, ground, and scatter maps which directly correspond to hypotheses on object type. Then 2D and 3D shape analysis are carried out to verify the object hypotheses. Our experimental results show that the new method performs well providing an autonomous system with surrounding 3D information and object classification.