Research On Stereo Matching Algorithm Based On Regional Hierarchy And Multi-scale

The objective of stereo matching is to obtain the distance information according to the matching points. Stereo matching is the basis step for the 3D representation, and the most important part in the stereo vision system. It is one of the hardest issues to solve thoroughly for the following reasons, the matching accuracy and the efficiency, the contradiction between the accuracy and the comprehensive of the disparity. The mainly contributions of the thesis are as follows:First, the basic theory framework of the stereo vision and the key technique are investigated. The theory and the algorithm of the binocular stereo matching are presentation in detail, which composing the mathematical basis of stereo matching, including the imaging geometry model, the constraint condition, the similarity measure and the algorithm evaluation standard. The two types matching algorithm of local matching and global matching are also analyzed and compared systematically.Then, the traditional regional algorithm is analyzed and the disadvantage of the SAD method from two aspects of theory and experiment is indicated. The improved hierarchical stereo matching algorithm based on the regional matching is presented. The color image is used in the algorithm and the size of the matching window is set. The updated threshold to precisely measure the matching points in accordance with the gradually increasing disparity is used. The false matching points are removed by the morphological filtering based on the space constraints simultaneously.Third, the wavelet transform and the multi-scale theory are introduced. A new improved algorithm based on the front method is presented. The direction selective constraint of the Gabor is used in this method. The disparity map obtained by the method is much more smooth and accurate for the more complete matching constraints. The hierarchical matching process is clearly reflected by the image which composed by the state values of the pixels in the way of gradually increasing disparity. Compared with the traditional way of finding the matching points, the reverse way in the method which obtaining the points in accordance with the disparity is novel.In summary, some new methods and idea which analyzed both in theory and experiment with good performance are presented in this paper. It is an important supplement of the stereo matching.