A Contour-Based Method for Whole-Shape Classification of 3D Point Clouds

With the wide application of 3D point cloud data in 3D modeling, scene reconstruction, virtual reality and augmented reality, how to carry out effective feature recognition of point cloud data has become an important topic in research. This paper proposes a novel method for whole-shape classification of 3D point clouds based on contour analysis of three orthogonal projections. The approach projects the 3D point cloud onto the XY, XZ, and YZ planes, uses the Alpha Shape algorithm to extract and smooth projected contours, and classifies segments as straight or curved via curvature analysis. By applying shape-fitting techniques and quantifying the ratio of curved to straight segments, the method categorizes models as planar surfaces, curved surfaces, or planar–curved mixtures. Experiments on the ModelNet40 dataset demonstrate reliable classification results, outperforming Efficient RANSAC for whole-shape categorization. The method is applicable to arbitrary models and supports preprocessing tasks such as data balancing and shape-aware feature learning.