Abstract:

The quality of point cloud registration directly affects the quality of 3D reconstruction. To solve the problem that the traditional K-4PC is time-consuming and prone to mismatching, a coarse point cloud registration method based on boundary centroid is proposed. By extracting the boundary of the point cloud, the surface features of the point cloud are preserved and the size of the point cloud data is reduced, which improves the speed of coarse registration. In order to speed up the extraction of boundary points, the K-D tree algorithm is used to search for k nearest neighbors. By registering the centroid of the boundary points, the initial distance of the point cloud is reduced and the degree of overlap is increased, ensuring the accuracy of coarse registration. The experimental results show that the proposed method is better than the traditional K-4PCS algorithm in terms of speed and accuracy. The speed of this method is about twice that of traditional K-4PCS. Both the translation and rotation accuracy are 40% higher than that of traditional K-4PCS. The proposed method has certain reference value for improving the speed and accuracy of point cloud coarse registration.

Key words: point cloud registration, coarse registration, fast registration, boundary extraction, k-nearest neighbors, boundary centroid, K-4PCS, K-D tree