Stereo Matching Research

Stereo matching, which is an important research direction in computer vision field,is a bottle-neck problem in the research of stereo vision. It's a great challenge to investigate a robust, precise, stable, and applicable stereo matching algorithm, which is both theoretically and practically valuable.In the thesis, we investigated a number of important problems related to stereo matching problem.The main work can be summarized as follows:1. Corner Detection Technique: We made deep study on the corners detection algorithm based on the gray-level image, then we proposed an enhanced MIC algorithm,which introduced into the median filtering,JND curve and the measure operators of texture technic. Experiment results showed that the stability,speed and the ability of MIC algorithm are improved obviously.2. Feature Matching Technique: We enhance the Zhang,s features matching algorithm. In the algorithm, we use 7-points algorithm instead of 8-points algorithm and self-adapted calculate the sample count. This two technique improved the stability and the convergence speed of Zhang,s algorithm. An iterative corners correspondence algorithm based on Multi-Neighborhood support technology is also proposed in this paper. First we use the five restrictions to the stereo matching process introduced by Marr and Frisby to restrict the searching area, then we brings forward an advanced method to measure the comparability of point features based on Multi-Neighborhood support technology; Last we introduced into the up/down and left/right symmetry test processes and iteration technology to improve the correspondence accuracy. These techniques can solve the problem that the traditional algorithm has bad real-time performance and low accuracy. The experiment results demonstrate that this algorithm is an speedy,stable and practical corners correspondence algorithm.3. Epipolar Rectification Technique: An enhanced epipolar rectification algorithm based on polar coordinates transformation technique is proposed in this paper. This algorithm first uses linear transformation to map the two epipolars to the same coordinate, then the rectification method which is based on polar coordinates transformation technique is used to complete the rectification process.This algorithm not only computes speedly,but also reduces pixel loss because of the difference sampling ratio.Experiment results demonstrated that this algorithm is an practical epipolar rectification algorithm.