Segmentation of 3D triangulated data points using edges constructed with a C1 discontinuous surface fitting

Current optic techniques make it possible to digitise objects or scenes in clouds of thousands of points. A large number of papers have been proposed to generate a polygonal representation associated to this digitised representation. For a better use in geometric and dimensional control or in reverse engineering such clouds need to be split in subsets that represent elementary surfaces. This paper presents a method to partition a polygonal network along the edges of a surface. Firstly, the areas with high curvatures are detected. In these areas, the position of the edge start is precisely calculated by locally fitting a surface that has discontinuous tangent. We call this surface 'absoid'. The whole edge is then calculated step-by-step by a new absoid fitting in the edge extension. Two absoid models are used, one has a sharp edge, the other has a rounded edge. After computation of all the edges, these edges are joined. Secondly, the facet network is deformed in order to make the facet edges and the calculated edges correspond. The moved facet edges are identified as border edges. The segmentation of the network is then obtained with a region-growing process. (C) 2004 Elsevier Ltd. All rights reserved.