Image-based Stereo Vision Research

3D reconstruction is a hot issue on the field of computer vision, and the stereo visionreconstruction method gets more and more researchers’ attention. In the real life,3D modelsare widely used in various fields, such as virtual reality, video games, the automation ofindustrial manufacturing, film and cartoon manufacture, pattern recognition, protection andrecovery of the cultural relics. The three-dimensional reconstruction techniques and theresulting technological innovation are changing every people’s lives, how to improve thealgorithm in the reconstruction process to be simple and reliable, has great significance.The mutual restraints among the drawing speed, quality, complexity in the traditionalreconstruction approach make its implementation is not very efficiently. Image-based3Dreconstruction is based on the image or image sequence to recover the three-dimensionalmodel of the scene, it costs less than traditional methods, has higher degree of automation,and broad application prospects. Which stereoscopic vision method achieves the3D depthinformation through the imaging and geometric principles, it imitates the mechanism ofhuman visual system to determine the depth of the object, is an important branch of the fieldof machine vision.In this paper, the method based on the stereo vision is researched deeply, to explore amore efficient and convenient method of modeling. The paper classifys the existing cameracalibration method first, and make the comparison tests on three extensive methods, finallythe planar pattern calibration method is chosen for the follow-up operation. And the paplecarrys out a detailed principle parsing on the Graph-Cut optimization algorithm, and applys itto stereo matching, provides a example for how to solve practical problems using with graphcut. The similarity between the grid projection points is used to determine the depth of thesurface, but some information noises arise, the result is not good enough specially on thediscontinuous regions. A method based on graph cut optimization can obtain the depthinformation of the object surface directly and simply. The depth information obtained throughthe graph cut optimization is more continuous. In the coordinate transformation process, whenthe two image pixel coordinates which projection from the same grid coordinate are twonon-integer, using bilinear interpolation approach to handle the pixel information.Accordingto the differences of the gray information, the energy function about the depth label isestablished, and using the graph cut theory to obtain the depth information of the surface.