A Laser Point Cloud Clustering Algorithm for Robot Navigation

A clustering algorithm for robot navigation and environment understanding is proposed.It is designed to deal with anisotropic distribution point cloud.This algorithm performs clustering according to the variation of density and spatial distribution of points.It combines concepts of information clustering with traditional DBSCAN algorithm.On one hand it keeps antinoise ability,and on the other hand it improves the clustering result by incorporating spatial probability distribution of point cloud.The algorithm uses an adaptive online parameter computing method to conquer the disadvantage of constant global parameter.Experiments on real data set validate that the proposed algorithm can separate connected objects where point cloud has similar density but different spacial distribution,and it can deal with point clouds with high noise.