Study Of Binocular Vision Based On Corresponding Points

Binocular vision research is one of the most active branches of computer vision technology, including many research aspects such as camera calibration,image acquisition, feature extraction, stereo matching, depth measurement and depth information interpolation. Among these research aspects feature extraction and stereo matching are the core parts of the binocular vision research. Therefore, this thesis has taken these two aspects as the key research contents, mainly containing:Firstly, based on dual CCD camera, the emphasis is stressed on the discussion of several coordinate systems in common use and the transition relations among them in this thesis.Secondly, concerning the feature extraction of image, the corner detection algorithm based on Freeman chain and the one based on area ratio are studied comparatively and an improved corner detection algorithm based on area ratio is proposed. The experiment results improve that the real-time characteristic of this improved algorithm has been enhanced under the condition of the same corner detection effect.Thirdly, the stereo matching is achieved by use of the similarity of two images calculated throught the correlation algorithm in this thesis. And then, taking the corners of an object as the corresponding points and considering of the epipolar restriction of corresponding points, an effective method is proposed used to match corresponding points. On the basis of such a result, the depth information is computed by using the disparity of corresponding points in two images.