Research On Stereo Matching Technology In Binocular Stereo Vision

Computer vision is a technology that how to imitate human vision with computer. In this field, binocular stereo vision has always been the hotspot of the research and it is widely used in target tracking, unmanned aerial vehicle navigation and other fields. Stereo matching is the key point in binocular stereo vision. The matching accuracy and speed can affect the application effect of entire system directly, therefore a more accurate and more rapid matching algorithm is one of the hotspots of current research.After researching a large number of domestic and overseas stereo matching algorithms, two matching algorithms are proposed from different research views. Meanwhile, in order to prove the research is practical, 3D reconstruction is accomplished by using the obtained disparity map. The main research works are as follows:1. The basic theory of stereo matching is researched and some common matching algorithms are analyzed and realized by programming,such as fixed window algorithm, shiftable window algorithm, multiple window algorithm and variable window algorithm.2. Some stereo matching algorithms can not have matching accuracy and matching efficiency simultaneously, aiming at this difficulty, a weighted-sparse region-based matching algorithm is proposed. Firstly, the gaussian function which similar to HVS(Human Visual System) is used to arrange different weights for different neighborhood points. Secondly, the sparse support window is used to aggregate matching cost to reduce computational time. Finally, the disparity map is post-processed to update the disparity of the occluded points and reduce the error matching points.3. Aiming at the problem that only one similarity measure is used as matching cost would lead to low matching accuracy and the time-consuming is affected by the size of support window, a local stereo matching algorithm based on the combination of multiple similarity measures is proposed. Firstly, four similarity measures which can complement each other's advantages are combined to form a new matching cost. Secondly, the guided filter is used to aggregate matching cost in order to obtain disparity map. Finally, the confidence estimation and left and right consistency check is used to further optimize the disparity map.4. According to the triangulation method, the 3D reconstruction is accomplished by using the obtained disparity map. In order to be convenient for the follow-up research, some stereo matching algorithms introduced in this dissertation and the two proposed stereo matching algorithms are integrated into software with a GUI(Graphical User Interface).