Automatic 3D point clouds registration method

3D point clouds registration is a crucial problem in reverse engineering. In order to register point clouds without manual information on objects, a novel method is proposed based on geometric properties of point clouds. The method consists four parts: selecting public areas, computing normal vectors and curvatures, finding corresponding points and further precise registration. Public areas of point clouds are selected through a scan order method. Normal vector and curvature of each point are calculated through surface fitting. Curvatures of points are taken as the registration relationship and all the pair-wise points with the same or similar curvature are extracted. The property of distance invariance in rigid body transformation is used to match the pair-wise points. Matching points are picked out and mismatched ones are excluded by comparing the curvatures in the k-neighborhood. After that the algorithm of quaternion is used to compute transform matrix and ICP algorithm is introduced to improve registration precision. Experimental results show that the proposed method is robust when registering point clouds of different scans.