Point cloud registration algorithm based on octree and 3D-SIFT feature extraction

The 3D detection of topological configuration components of additive manufacturing is crucial in post-processing. Topological configuration components have irregular surface geometry and a large number of holes distributed on the structure. Problems such as low registration efficiency and mismatching may occur when registering topological configuration component point clouds with existing 3D detection methods. A point cloud registration algorithm based on octree and 3D-SIFT feature extraction is proposed. Firstly, the point cloud data is preprocessed by statistical filtering, and then the feature points are extracted by combining octree sampling and 3D-SIFT to preserve the original features of the point cloud to the greatest extent. The relative dissimilarity vector is introduced to improve SAC-IA rough registration algorithm, and the point pairs with large differences are excluded in advance. Finally, an improved ICP algorithm combining two-way KD-tree and Welsch function is used for fine registration. In the Stanford University public point cloud registration experiment, compared with RANSAC+ICP, SAC-IA+ICP, ISS+3DSC+NDT and other algorithms, the average error is reduced by about 84.8%, 69.5% and 54.7%, respectively, and the registration time is reduced by about 86.7%, 78.1% and 58.2%, respectively. In the point cloud registration experiment of scanning topological structures, the root-mean-square error is 0.0612 mm and the registration time is 7.97 s. The experimental results show that the proposed algorithm can effectively improve the accuracy and efficiency of high point cloud registration, and is suitable for point cloud registration of topological configuration components with large data volume and complex structure, which lays a foundation for improving the post-processing effect.