Implementation On The Technology Of Multiview Reconstruction Based On Binocular Stereo Vision

The problem of 3d reconstruction is the research highlight of virtual reality and 3d visual field. This thesis mainly researches the implementation technique of 3D point cloud. With the basic principle of binocular stereo vision, construct a binocular stereo vision experiment system by multiple CMOS camera, developing the camera calibration, matching and point clouds data research, processing, etc.This thesis mainly research work are as follows:(1) The thesis determines the method of camera calibration based on 2D planar squares through the comparison of several camera calibration techniques. On this basis, the thesis constructs our experimental platform to make the cameras around the object and shoots image, calibrates internal parameters of each camera and calculates the structure parameters between every two cameras.(2) After camera calibration, every two cameras'perspective projection matrixes are relative to different world coordinate systems. Firstly the thesis uses the relative rotation between two cameras to calculate the unified camera rotation relative to the unified the world coordinate system. Secondly calculates the unified translation using unified rotation. Then every two images rectify epipolar to have parallel line. Finally harris corner detection as the seeds to regional growth algorithm to form disparity image.(3) On the basis of pairwise disparity image, the thesis forms 3d coordinates between the matching points through the triangle method. Combine the results, we have the high-density cluttered point cloud. The thesis designs and implements the octree and the quadtree sampling algorithms to decrease point cloud. Finally remove the mistakes on the surface of point cloud through three noise degrees.