Refined Keypoint Extraction Method for High-Density Point Cloud Objects Registration

The main objective of this paper is to address the problem of point cloud registration failure caused by the intricate spatial organization of point cloud data and the overlapping of sampling points throughout the registration procedure. We propose an effective approach to address point cloud registration problem, namely RP-ISS (Refined Point Cloud Intrinsic Shape Signatures). Firstly, the employment of adaptive scaling of the search space during keypoint extraction is utilized to enhance the effectiveness of keypoint extraction. The method effectively improves the stability and distinctiveness of the keypoint cloud. Subsequently, the implementation of fast point feature histograms are used to provide feature descriptions for the keypoints. Ultimately, the utilization of point cloud registration generates the acquisition of positional data pertaining to the target point cloud. The keypoint extraction technology tracks the actual world more effectively than others, making it able to endure in numerous environments. We use the ModelNet40 dataset to evaluate our suggested strategies, and we chose six distinct sorts of point cloud models for comparison. Extensive testing of the proposed method reveals that it is able to successfully and precisely extract superior point clouds for registration.