Feature Line Detection from Point Cloud Based on Signed Surface Variation and Region Segmentation

Feature line detection is important for the representing and understanding of 3D models. In this paper the concept of signed surface variation(SSV) is proposed. Except for the ability to represent local surface variation, SSV can also distinguish concavo surfaces from convex ones, so it is a good approximation to surface curvature. Based on SSV and feature region segmentation, a novel point cloud feature line detection algorithm is present. Firstly, points with large absolute SSV are recognized as potential feature points; Then they are segmented to different regions with the guidance of SSV; On the next, local mesh surface of each region is reconstructed from which boundary points are recognized and iteratively thinned using bilateral filtering algorithm; Finally, feature lines are linked by constructing the minimal spanning tree of thinned boundary points. Experiments indicate that our algorithm can recognize and segment potential feature points correctly, and can extract accurate feature lines completely.