Optimum Disparity Continuing Block Algorithm Of Stereo Matching

The correspondence matching of stereo reference images is vital to computer stereo vision processing. Though with good mathematics foundations, systematic theory and many algorithms well developed, current matching algorithms have many severity faults, of which both accuracy and efficiency are required to be improved. A novel matching approach is introduced here, in which the dense stereo matching algorithm follows a sparse matching process, estimating the essential parameters needed in dense stereo matching. Both the sparse matching and dense matching algorithm developed in the thesis possess a common feature, not like many previous algorithms which compared the whole lines or whole blocks of area to evaluate their correspondences, these two methods find out all of the possible matched pairs whose correspondences are precisely defined, then kick out false pairs, reserve optimum pair set as the matching result. The sparse matching uses a minimum topology value algorithm, which first obtains all possible vice matching line segment pairs by the direction along the line and magnitude of the line after the lines are extracted from the reference image pairs, then selects the final segment pairs referring to the topology value, basing the fact that matching pairs should locate similar positions in respective image. The matching result can be used to estimate the maximum and minimum disparity, gray correlation threshold and some other parameters which play important roles in dense matching. The dense matching used here is an optimum disparity continuing block algorithm, which search the disparity space by the means of scanning each horizontal slices to gather disparity continuing lines, then aligns such lines to form the disparity continuing block set. An optimum subset of it is reserved as the dense matching result. To present the algorithm visually and exactly, a lot of graphs and formal expressions are used, as well as many related algorithm flow.