PART-TO-PART POINT CLOUD REGISTRATION BASED ON LONG-RANGE REGION FEATURE EXTRACTION

Since the point cloud collection process is easily interfered with the environment, collected point clouds are often sparse and unevenly distributed, which makes it difficult to register between two set of point clouds. In this paper, we propose a partial-to-partial point cloud registration method based on end-to-end deep learning. We designed a residual curve attention module to extract long range regional features, and we proposed a key matching prediction module to generate key matching pairs. We used SVD method to transform the key matching pairs into the transformation matrix. Extensive experiments conducted on the ModelNet40 dataset manifest that our proposed method can achieve state-of-the-art performance, and the results on the KITTI dataset verify the effectiveness of our method in real-world environment.