Study On The Key Issues Of Three Dimension Reconstruction Based On Stereo Vision

With the development of3D technology,3D reconstruction technology,which isto restore the depth information of an object by two or more than two2-dimensionalimages,becomes more and more salient to the computer vision. And3D model thatmakes use of two2-dimensional images to restore the depth information of an objectwas widely used. This paper takes advantage of a set of parallel binocular camerasystem to shoot objects, and then restore the3D model of objects by the steps ofcorner detection, feature matching, and3D reconstruction etc. The binocular stereovision is widely used in many fields such as appearance design of industrial products,multimedia education, machine vision, digital entertainment, virtual reality, sculptureand architecture etc.First of all, on account of that video camera calibration is the premise in theexperiment that applies video camera, calibration quality determines the precision oflater experiments in a great degree. Hence, it is important to calibrate the videocamera before we conduct an experiment. This paper proposes a binocular visioncalibration method based on projection straight lines intersect to the binocular vision.By improving traditional calibrating system, the calibrating result has betterrobustness. Firstly, the left and right video cameras shoot a pair of images at the sametime, and then detect corners are detected via chessboard corner detection algorithm.Secondly, by covering the corner coordinates into linear equation, internal parameterand external parameter of video camera can be achieved.Secondly, a new algorithm which combines corner candidate screening and SIFTalgorithm is proposed based on the existing feature detection algorithm analysis.Firstly, corner candidate screening is carried out for the image of desired detectingcorner by applying8neighborhood similarity method. And3×3gray image is builtcentering on the candidate corner. Then SIFT detection algorithm is applied to detectcandidate corners for the built3×3gray image.Thirdly, after analysising and studying the principles and algorithms of cornermatching, a new improved matching algorithm is proposed based on NCC,thealgorithm which combines NCC matching principles and various constraints to findmatching points from the two images.Finally, a platform is designed including noise treatment, stereoscopic calibration,corner detection, feature matching, and3D reconstruction. And on that platform theproposed3-dimensional reconstruction is corried.