Experimental Research On Key Problems Of 3D Reconstruction In Binocular Vision

With the high-speed development of computer science and technology, computer vision has become an indispensable part of life, industry, military and other fields.Binocular vision, as an important part of computer vision, is to take the same scene from two different angles, and get the 3D information of space points by combining the image disparity and triangulation measurement principle, whose principle is to simulate the human visual system. This technology has the advantages of strong adaptability, wide implementation and low cost, which has become an important technical mean in the field of 3D reconstruction. Aiming at improving the accuracy of camera calibration and stereo matching effectively, so as to improve the efficiency of 3D reconstruction. Through accessing to relevant literature and information, in this paper, the following aspects of the work were studied:Firstly, the paper analyzed the Zhang Zhengyou plane calibration method, then a practical method of camera calibration based on an isosceles right triangle planar pattern was presented. This pattern has the advantages of simple manufacture, and does not need to know the length of the triangle. The method uses the improved Harris corner detection method to automatically extract the triangle vertices, and the pattern can be arbitrarily moved and rotated in the camera field of view. Then this paper used the two methods to calibrate the camera and analyzed the results.Secondly, this paper did a comparative analysis and experimental comparison on the image preprocessing of image gray-scale transformation, image filtering, histogram enhancement, and used the median filtering to deal with the noise of image, which laid the foundation for the follow-up work.Thirdly, with the problems of existing local stereo matching algorithm, which are low accuracy in weak texture region and disparity discontinuous region and image edge. A novel stereo matching algorithm based on the combination of matching costs is presented here. This algorithm can adaptively change the weights of the two parts according to the image feature, to effectively play the characteristics of each part. Moreover this paper usesthe semi-global matching algorithm to carry out optimized processing on disparity image.Higher matching accuracy is got.Finally, this paper used the camera calibration and stereo matching results and combined with the triangulation measurement principle to realize the 3D reconstruction. A dense disparity map and a higher accuracy depth map were obtained.