Integrating edge and surface information for range image segmentation

Abstract

Most conventional range image segmentation algorithms could be classified as either surface-based or edge-based depending on whether they extract the parameters of the surface or the surface discontinuities. Thus the processes of extracting the surface parameters and the surface discontinuities are treated as distinct from each other. In this paper we show that the synergetic combination of both surface- and edge-based segmentation processes can lead to more accurate segmentation results than just either of them alone. A segmentation algorithm is proposed which integrates edge and surface information in order to come up with a high-level description of a scene given a range image. Although the segmentation technique at this time is limited to polyhedral objects, it can be extended to include objects with parametric curved surfaces of quadratic order such as cones, cylinders and spheres that lead to parametric curves of discontinuity that are conic sections such as circles, ellipses and parabolas.