Self-localization Using Point Cloud Matching at the Object Level in Outdoor Environment

This paper presents a method for the localization of automatic navigation robot using accurate 3D point cloud matching. The method firstly applies a point cloud segmentation approach to get point cloud clusters, then associates different clusters of points between scans, which takes into account the attributes of point cloud clusters like the shape, the location and the point number, after that the velocity of these clusters is computed and dynamic objects are removed. Then the method applies a modified version of Iterative Closest Point (ICP) algorithm for accurate matching, in which the search for points between different scans is limited to associated static clusters. Unlike other localization method using point cloud matching, which fails to get accurate alignment in outdoor large scenes, especially when the distance between scans increases. The method in this paper is proved to be robust to long distance in outdoor environment, and the experiments show better precision of localization compared with the normal Generalized Iterative Closest Point (GICP) algorithm.