Research on three-stage point cloud registration mode optimized by simulated annealing algorithm

In order to solve the defects of traditional point cloud registration iterative nearest point (ICP) algorithm in the case of random initial rotation and translation matrix, it is easy to fall into local optimal, time-consuming iteration, and lacks self-adjustment ability, and meet the real-time requirements of using lidar to collect vehicle contour information. A three-stage point cloud registration mode optimized by annealing algorithm is proposed. Through the three-stage point cloud registration mode, the first stage is point-to-point registration, and the simulated annealing algorithm is used to select the matching pairs of feature points. The second stage registration is the registration of features. The first stage registration is used as input to construct feature point clouds for registration. The third part is point cloud to point cloud registration, and the ICP algorithm of KD-tree search optimization is used for the overall registration. Experimental results show that. This algorithm can obtain better vehicle contour point cloud, and by comparing classical ICP, principal component analysis (PCA)-ICP, FPFH-ICP and other algorithms, the algorithm proposed in this paper takes less time to register different environments and objects. The complete car contour can provide the basis for the control of the subsequent robot and has reference significance for the study of laser point cloud processing.