Abstract: In the traditional ICP algorithm, it is necessary to have a good initial position of two points cloud, otherwise it is easy to fall into local optimization when it is on time. Aiming at this problem, this paper proposes a rough registration method based on feature point extraction and pairing to adjust the initial position of the overlapping parts of two points cloud. Firstly, the feature points of the common part of two point clouds are extracted by using SIFT algorithm, and then the feature points of two point clouds are paired according to the Euclidean distance between the feature point method vectors, and the feature point pairs are purified by using the angle of the method vector. Finally, the rotation and translation matrices are solved by means of unit four yuan number method, and the coarse registration is completed. Experiments show that the coarse registration method based on feature point vector can provide a good initial position for the precision registration, and can avoid the phenomenon that the local optimization is caught on time to a certain extent.

Key words: SIFT algorithm, method vector European distance, method vector angle, unit four yuan, ICP algorithm