The Segmentation of A Point Cloud Using Locally Fitted Surfaces

This paper proposes a new method for the segmentation of the scanned point clouds that achieves an accurate definition of objects' faces, critical for their digitalization. The topological structuring of the input point cloud is realised by a nearest-neighbour graph, while locally fitted surfaces are used to quantify the objects' shapes. Noise is filtered by a graph-cut and the segmentation is performed by linking the nodes to those faces to which they fit best. The results show that the proposed method is capable of improving the accuracy of the segmentation by 5%, while reducing the computation time in comparison to the state-of-the-art.