Research on algorithm of point cloud coarse registration

Targeting at prerequisites in the process of registering 3D point cloud data under different angles of view by ICP algorithm, a new automatic coarse registration algorithm is presented in this paper. Firstly, through computing normal vector variation of point cloud within different radius, simplify point cloud and keep necessary key points. Then comprehensively use three geometric features of key points in point clouds to search for matching point pairs. Next, use ZNCC (zero-mean normalized cross-correlation coefficient) registration criterion which is used in 2D image matching, so as to compute neighborhood similarity between matching point pairs. According to neighborhood similarity, build one-to-one correspondence between point pairs and establish coarse registration point pairs afterwards and finally get coarse matching parameters. The experimental results show that either used alone or combined with ICP algorithm, this algorithm is a great improvement in comparison to solely using ICP algorithm or other improved ICP algorithms in terms of efficiency and accuracy.