Adaptive Support-Weight Approach For Stereo Matching In Binocular Vision

Binocular vision is an important research field of computer vision, the purpose of binocular vision is to recover the3d information of a scene from the2d images information by simulating the human eye vision system. Binocular stereo vision consists of four steps: image capturing, camera calibration, stereo matching and3d reconstruction, where the stereo matching is the key technology. Although there are a lot of stereo matching methods, but there are still many problems in practical application. The purpose of this project is to find a fast and accurate dense stereo matching method for the naked eye3d video processing system of Temasek Polytechnic school. Global methods can obtain high precision results, but their computation structure is complicated, and not easy for hardware implementation. Local methods have simpler computation structure, and easy for hardware implementation, however the matching precision of the results is also lower. As Yoon introduced the Adaptive Support-Weight approach, the performance of local stereo matching methods has recently experienced great improvement, even are better than some of the global methods, but the Adaptive Support-Weight approach still have two important questions:how to choose the proper parameters, and the computing time is longer than other local methods. This paper is to explore a kind of local stereo matching method which has high computing speed and good performance.The main work and research results are as follows:1、The matching model of binocular stereo vision and the existing stereo matching methods are been briefly analyzed, the comparison of the six kinds of correlation measure function which are commonly used in stereo matching are presented, the stereo evaluation standard and the occlusion detection and filling post-processing measure have been given, the performance of different correlation measure functions have been compared, the effectiveness of the post-processing measure has been verified.2、The Yoon Adaptive Support-Weight approach has been studied in detail, the gestalt principles and the implementation method of Adaptive Support-Weight have been produced, the parameter problems of the Adaptive Support-Weight approach have been pointed out, and then a Relational Analysis Method is been presented, which discussed the parameters of the Adaptive Support-Weight approach by considering the changes the different parameters in a time. Finally, the experiments show that the different combinations of different parameters have important influences on the Adaptive Support-Weight approach.3、A kind of modified Adaptive Support-Weight approach has been proposed. This paper improved the Adaptive Support-Weight approach from two sides:computation speed and the matching performance. In order to improve the computation speed, the RGB color space has been used to take place of the CIELab color space when computing the color similarity, the single support-weight has been used to take place of the double support-weight when aggregating the raw matching cost, and in stereo matching using a small support window instead of a big support window. In order to improve the matching performance, in the modified method, the correlation measure function with better matching performance has been used, and at the end of the traditional matching, a kind of disparity refinement method based on the adaptive support-weight approach has been brought in. At last, experimental results improved that the modified method is effective and excellent.