Efficient Contour Extraction in Range Image Segmentation for Building Modelling

Abstract

This paper presents a new technique to solve the contour and region extraction problems that appear in the edge-based segmentation of range images. A different approach is used in relation with previous work where the enclosed surface information is considered. A fast technique only based on the edge information is proposed. It generates a closed boundary representation from a binary edge map as input and consists of four stages. In the first stage, the points from the binary map, points belonging to crease and jump edges of the range image, are triangulated through a 2D Delaunay algorithm. Then, the obtained triangular mesh is considered to be a weighted graph, where each node in that mesh is a node in the graph and the edges of the triangles are considered as edges of the graph. The 3D length of the triangle’s edges are the weight associated with the corresponding graph edge. In the second stage, the minimum spanning tree (MST) of that graph is determined. Next, a post-processing is responsible for removing some short branches generated by the MST. Finally, the regions contained in the range image are extracted by analysing the polylines that define the contours of the different regions. Experimental results with different range images are presented.