Point Cloud Registration Based on Global and Local Feature Fusion
Abstract
Global feature extraction and rigid body transformation estimation are two key steps in correspondences-free point cloud registration methods. Previous approaches only utilize the global information while the local information is ignored. Moreover, global and local information may play different roles on multiple point clouds. In this paper, we verify the sensitivity of different types of point clouds to global and local information. We conducted extensive experiments on the ModelNet40 dataset by the SGLF-DQNet. Through the experimental results, we summarize the point cloud structure of the sensitivity to global and local features in the correspondence-free point cloud registration task.
