A LiDAR Mapping System for Robot Navigation in Dynamic Environments

Simultaneous Localization and Mapping (SLAM) technology based on multi-line LiDAR enables real-time robot positioning and environment mapping in unknown environments. However, three-dimensional point cloud maps constructed in this way cannot be used directly for navigation. Meanwhile, during the mapping of complex dynamic scenes, the interference of dynamic objects will result in maps with artifacts, which exist in the form of obstacles in the map, causing adverse effects on the subsequent navigation. To solve these problems, we propose a LiDAR mapping system suitable for robot navigation in dynamic environment. Firstly, a LiDAR SLAM algorithm based on intensity scanning context (ISC) loop closure detection is proposed to construct the original 3D point cloud image, it can effectively suppress the pose drift of odometer, and ensure the real-time performance of the algorithm with the addition of loop closure detection module. Then we design a method to obtain a two-dimensional grid map suitable for robot navigation, after filtering false obstacles caused by dynamic object trajectories in the original point cloud map, a composite filter is used to further filter the point cloud map to avoid the influence of uneven ground point cloud on navigation, then a two-dimensional grid map is obtained through probabilistic updating. The experimental results show that the mapping system is suitable for global navigation, and the positioning accuracy of the proposed algorithm on KITTI data set is 46.5% higher than LeGO-LOAM, the average positioning accuracy is 59% higher than LIO-SAM, and the running time is increased by up to 0.42%.