Research On Local Stereo Matching Method Of Binocular Vision

Binocular vision is an important research field of computer vision, and it simulates the visual system of human being to obtain the3d information of a scene from the2d images, which is widely applied into many fields, such as driverless car navigation, three-dimensional measurement,3D film, industrial inspection and so on. Binocular vision includes several steps: image acquisition, camera calibration, image rectification, stereo matching and depth information extraction. Stereo matching is an important and difficult point, matching accuracy and calculation speed of the algorithm directly influence the performance of the binocular vision in the practical application.This paper introduces the related principle of binocular vision in detail, focuses on the local stereo matching algorithm which is easy to realize, and the matching cost functions and cost aggregation methods of local stereo matching algorithm are studied and compared, and on the basis of existing algorithms, the adaptive-weighted method is optimized. In order to improve the computation speed, the guided filter is brought in to calculate the weights, the computational complexity is reduced by using integral image method. In order to improve the matching accuracy, in the modified method, the matching cost function with better performance has been used, and a step of disparity calibration based on the adaptive-weighted method has been added to reduce the matching errors which existed in the initial disparity, the occlusion is handled to improve the matching effect by some post-processing measures such as left-right consistency check and occlusion filling. The improved algorithm uses Middlebury’s image pairs and rectified stereo pair of images to test respectively, and the result is evaluated and compared to improved that the modified method is effective and excellent.The project builds a binocular vision image acquisition platform, and through the MATLAB programming to obtain images synchronously in binocular camera. MATLAB Calibration Toolbox is used to obtain the intrinsic and external parameters of the both camera, and complete stereo calibration. In the VS2013programming environment, the calibration results are imported to OpenCV, the real scene image pairs are corrected, and undistorted rectified stereo pair of images are obtained, the disparity map which gets from stereo matching process by the proposed approach will be used to compute3d information of the scene, and the3d information measuring function of binocular vision can be achieved.