A Color Stereo Vision System for Extraction of 3D Edges of Obstacle

In this article, we present a fast and self-adaptive stereo vision system for road obstacle detection. In a first step, vertical edge points are extracted using the operator color declivity. Then, vertical edge points are associated using dynamic programming based on geometric, non-reversal, uniqueness and color photometric constraints. Wrong matchings are detected and corrected using constraint of disparity continuity applied on 3D shapes which have been previously constructed. Finally, 3D edges of obstacle are extracted. Performance of segmentation, matching and extraction of 3D edges of obstacle are discussed. Experimental results are shown