Point cloud registration method based on 3DMatch network and improved ISS algorithm

Abstract

Point cloud registration is a key technology in the fields of SLAM and 3D reconstruction. Aiming at the problem of the high initial position of the point cloud by the traditional ICP algorithm, and the problems of inconspicuous feature and low matching rate caused by the use of random sampling strategy in the classic 3D Match point cloud registration algorithm, the improved ISS key point extraction algorithm is used to obtain key points. And input the 3D Match network to obtain the 512-dimensional feature vector, RANSAC was used to eliminate mismatches, and then rigid body transformation matrix was calculated by SVD to complete coarse registration of point cloud. Then use the ICP algorithm to perform the fine registration of the point cloud. The final experimental results show that the algorithm has excellent performance and can complete the point cloud registration quickly and accurately.