Research on High Precision Point Cloud Registration Algorithm in Dynamic Environment

The localization technology based on Lidar is an important method to realize the precise localization and navigation of unmanned vehicle, and it is also the premise to realize the application of unmanned vehicle in more fields. In many application scenarios, dynamic objects may interfere with precise localization, resulting in degeneration of Lidar localization accuracy or even failure. This paper proposes an accurate point cloud registration algorithm for dynamic environment. Firstly, the point cloud is roughly registrated to obtain the rough pose estimation, and the potential dynamic object point cloud is segmented, then it is removed from the original point cloud. Finally, based on the corrected point cloud, the high-precision point cloud registration is performed with the pose obtained by the rough pose estimation as the initial pose, so as to obtain the accurate pose estimation of the unmanned vehicle. The method is applied to Lidar to realize the fast and high-precision lidar localization of Ackerman unmanned vehicle in dynamic environment. The algorithm is verified in the campus environment. The results show that the algorithm is reliable and effective in the dynamic environment, and is superior to the common point cloud registration algorithm.