Curvature-Adaptive Deformation Graph for 3D Point Cloud Non-Rigid Registration Under Multi-Geometric Constraints

Pose initialization and reliable correspondences are the keys to accurate 3D point cloud registration. In view of the problem that the existing non-rigid registration methods do not perform well in the face of large deformation and missing correspondences, a non-rigid registration algorithm based on 3D point cloud curvature adaptive deformation graph and multi-geometry pruning strategy is proposed. Firstly, by sampling the source point cloud based on the Gaussian curvature and local geodesic distance, a node graph reflecting the deformation of the source surface is constructed adaptively. The nodes can be evenly distributed on the source surface by sampling the key deformation points and controlling the sampling density. Secondly, the point cloud SHOT feature and curvature are used to find the initial correspondences, while the reliable correspondences are obtained according to the diffusion pruning technology. Finally, the correspondences are re-searched during the registration optimization, and the mismatches are removed to constrain the deformation domain according to the distance and normal pruning. The experimental results on the ANIM and Human-motion datasets show that the proposed algorithm can obtain a good initial pose and remove a large number of mismatches. While the average registration error is reduced by 1 to 5 times, the running speed of non-rigid registration is increased by 3 to 7 times.