Feature-point-based stereo correspondence from stereo images using geometric relational matching

Stereo vision is a technique frequently used for recovering the threedimensional structure of a scene from stereo images. The main aspect of stereo vision that is still under-developed is the correspondence problem. In this research, distinct features from a pair of stereo images are extracted using feature extractor. A windowbased feature points detector is then developed to detect feature points from extracted features in both images. These points become the basis for feature point-based stereo correspondence. The extracted feature points are then connected to form a straight-line segment in both images. A match function representing the geometric characteristics of straight-line segment and the requirements of the epipolar and disparity constraint of straight-line matching is formulated. The similarity measures are applied to decide whether straight-line segments, one in the left image and the corresponding one in the right image, are potential matches.;A Geometric relational matching method uses three and five straight-lines from left images to construct model polygons. Then the right image is searched by an exhaustive search method to find matching polygon by comparing geometric similarity of each straight-line. For a reliability of matched polygons, a polygon grouping mechanism selects three matched polygons from the left image and corresponding one in the right image. The center of each polygon is found and connected to form a new polygon in both images. Similarity measures of each polygon are then compared using geometric relational matching method.