The Key Technologies’ Rresearch On3D Reconstruction Based On Binocular Vision

The key technologies’research on3D reconstruction based on binocular vision, which is one of the main contents of the computer vision, researches the ways of reconstructing the object by being taken into two or more images, especially the techniques of binocular vision reconstruction based on two images. The key technologies’research on3D reconstruction based on binocular vison has important theoretical significance and application value.Based on analysis the related technology and research of many literatures, camera calibration, corner detection, stereo matching,3D visualization and other key technologies related to3D reconstruction are researched in this thesis. The corresponding work is shown as follows:(1) The Internal and external parameters of camera calibration parameters are obtained based on Tsai calibration method, and the calibration accuracy is improved by handing the camera’s radial distortion. According to experimenting different number of calibration images, the influence of the calibration image’s number to the camera calibration parameters and error are analyzed. This calibration method has excellent noise immunity proved by numerous experiments.(2) SUSAN corner detection algorithm and Harris corner detection algorithm in this thesis. According the merits and demerits of those two methods, the Harris algorithm is improved based on the combination of multi-scale and SUSAN algorithm. The improved Harris algorithm is proved to have Superiority and noise immunity through many test results.(3) The disadvantage of Harris algorithm is improved, for which can only achieve pixel level of precision. The quadratic surface fitting method and error iteration method are studied to apply the sub-pixel corner detection, with comparing the two methods through the experiment analysis, the quadratic surface fitting method is more suitable for this thesis.(4) According to the study the classic correlation matching method, the information entropy and HSV color moment is introduced into the corner matching. By combing the Harris corner detection and the advantage of the SIFT algorithm, Harris-SIFT method is proposed. The sample test method is applied to calculate the matching veracity of the matching algorithm. Test results proved the effectiveness of Harris-SIFT(5) The combination of matching corner and stereo camera models is used to get three-dimensional coordinate point to do3D reconstruction by vertex list and index list on the platform of DirectX.