Research On3D Model Reconstruction Method Based On Binocular Stereo Vision

3D reconstruction of objects is a common scientific issue and core technology in the fieldsof computer aided geometric design, computer graphics, computer animation and digitalmedia creation.3D reconstruction based on binocular stereo vision shoots the same sceneusing two cameras in different positions or a camera by moving or rotating, calculates thedisparity of3D points in two images, obtains the3D coordinates of the point and reconstructsthe scene. The technology has become an important method in the field of3D reconstructiondue to its advantages of low cost, wide applicability. How to calibrate the camera and rectifythe images, how to improve the accuracy and efficiency of stereo matching, and reconstructthe point cloud efficiently are difficulties and focus. In this dissertation, the above four keyaspects are studied.Firstly, the calibration methods of two camera imaging models are analyzed. According tocharacteristics of the binocular camera and scaffold used in the subject research, thesecondary calibration method using2D checkerboard plane template is proposed to improvethe calibration accuracy.Secondly，a rectification method for uncalibrated images based on internal parametersestimation is proposed, only several groups of the matching points in the original images arenecessary, the matching points in the rectified images satisfy the polar geometric constraints,minimizing the Sampson error values of all matching points and the system equation isobtained, the internal parameters are estimate and the optimal rectified images are gained.The experimental results show that we can get better rectification images when the calibrationparameters are unknown and provide standard stereo image pairs for stereo matching.Thirdly, a stereo matching algorithm based on combined similarity measurement andguided filter is presented. Pixel gray and gradient of gray are combined linearly as the similarity measurement, the guided filter, an edge-preserving filter and its calculation does notdependent on filter size, is used as adaptive support weight. Experimental results show thatthe algorithm can obtain effective disparity image and has high efficiency and good radiationinvariance.Finally, Aiming at the point cloud simplification in3D reconstruction, a simplificationalgorithm with preserved edge points based on octree is put forward, the huge scenery pointclouds are simplified, the edge points are preserved and the point clouds is reduced, so theefficiency of point cloud reconstruction is improved.