The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

Research Paper #574

Title:Computing Surface-Based Representations from Range Images
Authors:Trucco,E; Fisher,RB
Date: 1992
Presented:Submitted to IEEE INternational Symposium on Intelligent Control, August, 1992
Abstract:Visual perception is an essential sense for robots whose task is to learn about the environment. A crucial role is played by the representation of the world that the system must build and maintain, and by the algorithms adopted to compute the representation itself. In the recent years range vision and surface-based representations have been increasingly used for robotic applications. A surface-based representation of a scene is typically built from a range image by segmenting the image into a collection of surface patches, whose position, orientation and shape in 3-D space is estimated. W present a system which computes such a representation efficiently from range images. We consider patches belonging to a qualitative shape catalogue suggested by differential geometry as the main features of our representation. The surface is segmented and the patches classified by estimating the sign of mean and gaussian curvature at each nonsingular point. The parameters of the local differential structure of the surface are also estimated (augmented Darboux frame). We discuss shape distortion effects introduced by gaussian smoothing, modelled according to the diffusion paradigm. We propose a shape-preserving boundary condition for the diffusion equation. We pre-compute depth and orientation discontinuities maps and use them to restrict the diffusion process to non-discontinuity points, thus avoiding the creation of spurious curved regions around discontinuity contours. Some experimental results and quantitative tests are also given.

[Search These Pages] [DAI Home Page] [Comment]