Binocular Stereo Vision-based Stereo Matching Algorithm

Binocular vision imitates the human's eyes which observes objects from the two different viewpoints, and then gets the spatial information and the contour information under the help of some imaging geometry principles which can get a result by calculation, and then get the result of the subject's 3D information. Stereo matching, which is the core technology of the Three-dimensional reconstruction and also the focus study and a "pathological"problem of stereo vision, is the process of searching the same point in the imaging process of the pair of images and obtaining the deviation of images.Image preprocessing is the foundation of a binocular stereo matching algorithm that can be realized. In order to lay a solid foundation for stereo matching, this paper adopts the optimal algorithm for the image pretreatment of graying, smoothing, sharpening on VC platform.When studying on the stereo matching algorithm, this paper improved the area matching algorithm and feature matching algorithm. In order to reduce the search range and the rate of false matches, this paper improved a area matching algorithm of fixed windows and proposed a new area matching algorithm based on disparity gradient, the new area matching algorithm in this paper introduced a new similarity measure function instead of the traditional function of SAD, and based on the disparity gradient theory that using different window sizes in different regions to match images. To feature matching algorithm, an improved feature matching algorithm that based on Harris's corner with the features of scale-invariant was proposed, and this matching algorithm introduced a new detection operator with scale-invariant feature to improve the classical Harris's detection operator, and when the feature vectors is generated, the most stable performance descriptors is introduced to improve the stability of the matching. Finally this paper analyed the advantages and disadvantages of those two algorithms, and new matching algorithm combining region and feature is projected. In this algorithm, the corners is extractded by the improved Harris's detection operator from this paper in the areas of the edges which are decided by the principle of disparity gradient after extracting the edges of images, then the feature points are matched by feature matching algorithm. Finally the images that matched by feature matching algorithm are matched by the improved area matching algorithm which projected in this paper. So this algorithm not only can reduce the search ranges to improve the speed of the matching, but also can be result in the dense disparity map to improve the accuracy of matching. The algorithms matched with standard images in this paper were cmpared with two aspects of qualitative and quantitative, the results show that the correct rate of the matching algorithm which combined with area and feature is higher, and the performance is also more stable.