Neighborhood Segmentation Based Feature Vertex Detection for Point Cloud

Surface geometry feature detection plays a critical role in the representation and understanding of geometric objects. This study introduces a robust and reliable method to accurately capture the local surface fluctuations of point clouds across diverse feature scales. By defining a bilateral metric that integrates weighted centroid projection and weighted normal variation, we systematically quantify surface discrepancies. Leveraging this metric, we develop a structure-to-detail feature-sensitive algorithm that enables precise localization of geometric features at multiple scales. Extensive experiments conducted on heterogeneous point cloud datasets-including synthetic models, real-world scans, and noisy environments-validate the efficacy and universality of the proposed method. Computational results demonstrate that our approach can effectively identify a broad spectrum of geometric features, ranging from complex topological structures and textured surfaces to minute details, subtle curvature variations, and sharp geometric boundaries.