Research And Application Of Matching Algorithm Based On Binocular Stereo Vision

Stereo matching is an important field in computer vision, with many new methods and large numbers of new ideas, stereo matching technology has been rapid development and the technology has been applied to many fields, such as three-dimensional reconstruction, robot navigation, medical diagnostics and so on. The paper is based on the principle of stereo matching algorithm, the study based on the basis of dynamic programming optimization deepening disparity map and get more precise, more clear boundary parallax disparity map, And the application of research results achieved in practice. Research and work is as follows:Firstly, we introduce the camera model and a variety of camera calibration methods, and detailed analysis, and for the advantages and disadvantages of between traditional calibration methods and self-calibration method, using a new, flexible flat camera calibration method(Zhang Zhengyou plane calibration method) to complete camera calibration, this method has the advantage which the traditional methods and self-calibration method does not have.Secondly, the stereo matching algorithm is focused on the thesis, and propose research of stereo matching algorithm based on image segmentation and control point of this article, for the case which there is only bound by the scanning lines, there is no constraint between the lines in the calculation of the dynamic programming, and easy to produce "fringe flaws " phenomenon has been improved by adding five more stringent constraints in the dynamic programming to obtain ground control points to guide the selection of the path, and in a certain extent, this improves the accuracy of parallax. In parallax optimization, based on the thinking that the disparity change only appeared in the areas of the divided region and within the boundaries of segmented regions is smooth parallax, we use Mean-Shift image segmentation methods for image segmentation, and use fast ballot in each divided area optimization dynamic Programming resulting disparity map. This algorithm is derived by the disparity map not only improve precision, and clear parallax boundary, there is no "fringe defect" phenomenon.Finally, the algorithm is applied to practice, through a camera fixed on the industrial robot MOTOMAN-UP6 end to get the reality of two images,and then use stereo matching algorithms developed to complete stereo matching, and complete target extraction and analyze the optical axis parallel binocular vision system principles, using parallax distance method to complete three-dimensional reconstruction of spatial points.