Point Cloud Edge Extraction Based on 3D Point Separability Filter with Spherical Mask

In this paper, we propose 3D Point Separability Filter for edge and corner detection in point cloud data. Unlike existing approaches that measure deviation from planarity, our method compares the class separability of points between two regions along surface normals, enabling direct feature extraction without voxelization. This approach produces consistent feature localization where values reliably increase with the number of intersecting surfaces, enabling both edge and corner detection by adjusting the threshold value. Comparative analysis shows our method outperforms eigenvalue-based Surface Variation approach. Experiments confirm effectiveness in detecting geometric features even with outliers, noise perturbations, and varying point density.