To obtain depth information and a 3-D description of the objects in a robot work cell, a number of passive and active methods have been proposed. One is a stereo vision technique that uses two cameras at different viewpoints. The major issue in any stereo vision system is the correspondence problem. In this work, a multi-channel feature-based stereo vision technique is proposed in which curve segments are used as the feature primitives in the matching process. A relaxation technique is used to obtain a global match betwen the curve segments in the left and in the right image. Local information about the curve segments as well as the structural information about the objects in the scene are incorporated into the stereo technique. The specific goals are: 1. to develop the proposed stereo technique, and evaluate its performance on a large number of stereo images; 2. to integrate the stereo technique into an operational robot work cell; 3. to evaluate its operformance, and compare with existing techniques.
