Dynamic Object Separation and Removal in 3D Point Cloud Map Building

Abstract

This article aims to solve the problem of ghost trail effect left by dynamic targets in the process of mapping. We present a novel dynamic object removal approach, which is robust and efficient. Firstly, we rasterize the map and LiDAR scan key frame output by LIO-SAM, and then execute map division and preliminary screening of potential dynamic regions, finally carry out two steps clustering and dynamic weight update. In addition, we conduct the real experiments on the robot, and the experimental results prove that the average preservation rate of static points reached 90.51%, and the average rejection rate of dynamic points reached 97.36%. As verified on real experiment, our method can directly remove dynamic objects in large areas efficiently, it a good removal effect on low dynamic or temporarily staying dynamic objects.