Abstract: A multi-view point cloud registration method based on K-means+〖KG-*3〗+ is proposed for the possibility of noise, outliers and occlusion in large scale point sets. Firstly, the random seeding technique of K-means+〖KG-*3〗+ algorithm is used to select the initialized center of mass from the subsampled multi-view point sets, and the clustering is completed according to the basic principle of the algorithm. Secondly, the point cloud data are stored in the K-D tree structure, and the nearest neighbor search algorithm is used to establish the corresponding relationship between the point sets, so as to improve the search efficiency of the corresponding point sets. Finally, the rigid transformation parameters between the point cloud data obtained by the clustering of each view and all views are calculated according to the scanning sequence by the iterative closest point algorithm, and the errors caused by pairwise registration are evenly spread to each view until the final registration result is obtained. Experiments on Stanford 3D point cloud datasets show that the proposed method has higher registration accuracy and robustness than partial multi-view registration algorithms in recent years.

Key words: point cloud registration, multi-view registration, K-means+〖KG-*3〗+ algorithm, iterative closest point algorithm, rigid registration