Efficient registration based on global 3D motion vector for sequential point clouds

Recently, acceleration of registration for sequential point clouds is a major research. However, ICP algorithm requires long computation time generally, and correspondence based methods depend on accuracy of feature point matching. In this paper, we estimated a 3D motion vector including influence of rotations, applied it ICP and correspondences based method. According to experiments using actual data, we confirmed proposal methods were same accuracy and faster than ICP.