DXICP: A Fast Registration Algorithm for Point Cloud

Abstract

The registration of point cloud is essentially to obtain a relatively accurate coordinate transformation matrix through operation. Generally speaking, the registration is to discover the position conversion matrix of the overlap between clouds. In this paper, a new algorithm DXICP is raised to handle the problem of low precision and slow time of traditional ICP. First, uniform downsampling is carried out after the input of the original point cloud. Then the 3D-SIFT algorithm is used to extract the feature points, and the FPFH algorithm is used to describe that. Finally, the RANSAC algorithm is used for rough registration, and then the optimized ICP is used for precise registration. The algorithm in this paper is proved to be better than the traditional ICP in both accuracy and speed.