An improved iterative nearest point registration method for vehicle-mounted laser point cloud

Abstract

The laser point cloud is sparse and noisy, and the traditional iterative closest point (ICP) algorithm has poor robustness and long convergence time. In order to further improve the accuracy and robustness of point cloud registration, an iterative nearest point registration algorithm (FPFH-ICP) based on normal vector angle generation of Fast Point Feature Histograms (FPFH) is proposed. Firstly, the voxel grid filter and Statistical-Outlier-Remove filter are used for sampling, and the feature point normal vectors that meet the threshold conditions are screened to generate the point feature histogram. Then, the sample consensus initial aligment (SAC-IA) algorithm is used for initial registration, and the K-D tree accelerated iterative ICP algorithm is established to achieve fine registration. In this paper, multiple registration experiments are carried out on laser point cloud data with different characteristics in the two scenarios of straight and steering, and the results show that the improved FPFH-ICP can achieve efficient and robust registration for vehicle point clouds.