Research On 3D Reconstruction Based On Binocular Stereo Vision

With the rapid development of science and technology, computer vision as a multidisciplinary cross emerging subject, either in theoretical research or in practical application has got rapid development. In the field of computer vision, binocular vision technology always gets a lot of attention. The binocular camera is used to imitate the human eyes to collect the outside information. The information is entered into the computer to realize the modeling and simulation of the human brain, and then depth information of target object is obtained according to the triangular geometric principles. Thus, the 3D reconstruction has been done to achieve the three-dimensional recovery of the target scene and object.The 3D reconstruction based on binocular stereo vision has been deeply studied in the paper It is mainly divided into camera calibration, image preprocessing and external feature extraction, stereo matching and several important components of 3D reconstruction. Three parts of research have been done in the paper:(1) Two CCD camera calibrations have been completed in the terms of camera calibration. A conventional calibration method-Zhang Zhengyou checkerboard calibration method is used for camera calibration to combine with the project requirements, then calibration is realized via the MATLAB calibration box and the OPENCV library, the advantage and disadvantage of the two methods are compared under the same number of calibration board pictures.(2) During the process of extraction of image preprocessing and external characteristics, classical de-nosing methods is used to accomplish the image preprocessing, which can solve the problems of noise and distortion existing in image. In the extraction of image features, combined with the project sources, an improved active contour model algorithm is mainly adopted to extract the contour of the object in the image, and the contours of the target objects can be extracted effectively by the proposed algorithm.(3) During the process of stereo matching, the currently predominant matching algorithm, that is stereo matching algorithms based on sift and surf feature have been analyzed and compared, and an improved feature points based on sift algorithm is proposed. First, the sift feature extracting is used to extract basic feature points, these feature points are used as the starting points, and then the similar feature points around them are found to form the feature point block; second, the matching algorithm is used to match, then the mathematical morphology is analyzed, the disparity map is repaired to get the final disparity map. Through a comparative analysis of the experiment, the feature false match can be effectively reduced by the algorithm, and also the detail of objects disparity map is well expressed.Finally, the ideal parallel binocular stereo model and the reality parallel binocular stereo model are analyzed and studied by binocular stereo vision model in the paper. The system framework of the reconstruction and the construction process are both completed according to camera calibration, the image preprocessing and external feature characteristics, stereo matching. The depth chart and reconstruction renderings of its three-dimensional space are obtained by the construction.