Surface Extraction and Boundary Detection Based on DBSCAN Clustering in 3D Point Clouds

Surface extraction from point clouds has many applications in product design, reverse engineering, industrial manufacturing and more. One of the key challenges is the detection and segmentation of overlapping areas of different surfaces. Existing methods are mainly on local judgment of similarity features, which can lead to either over- or under-segmentation. The proposed method in this paper utilizes DBSCAN for 3D point clouds surface extraction and boundary detection. Prior to extraction, the normal vector at sharp features is estimated based on a voting method. The surface extraction process is comprises two stages. First, the local quadric surface is determined based on the candidate sample points, and the points belonging to the same local surface are clustered. Second, in the merging stage, where the local surfaces are combined into a single smooth surface using three merging conditions. To demonstrate its effectiveness, the proposed approach is accessed on synthetic and real datasets, showing that the method can address the problem of insufficient and excessive segmentation of curved point clouds, and achieve high extraction accuracy.