Exploiting image-plane data in the interpretation of edge-based binocular disparity

Abstract

We consider the exploitation of 2D, image-plane measurements by the 3D grouping (CONNECT) and contour segmentation/description (GDF) modules of the TINA binocular vision system. The combination of point strings obtained separately by CONNECT from matched and raw edges is discussed and a simple approach to grouping 2 and 3D edges presented. An upgraded GDF is also described. This uses raw edges to guide both segmentation and description. Over-segmentation is thus reduced and a greater proportion of the connected edges assigned to reliable, 3D descriptive elements. Should binocular correspondence fail for any reason, default, two-dimensional descriptions are produced automatically.