3-Dimensional Robot Vision Using Image Database

Abstract

This paper discusses image registration on the basis of the node structures which are represented by the arcs linked to them and their shapes of the image graph and its application to stereo visions (motion stereo and binocular stereo vision). It is necessary but difficult to find matching between feature points on a pair of stereo images when a vision system extracts three dimensional positions and shapes of objects. This paper proposes a matching manner using an image database in the case when the occlusion occurs on the pair of stereo images. Since our approach is based on an algebraic expression of edge line image using unit vectors and their concatenation operators, it is insensitive to scaling, rotation and gray level modification between the pair of stereo images. The matching process makes use of a degree of structural similarity between nodes of the image graph. The experimental results confirm that our approach is effective for the image registration between the pair of images including the occlusions.