Image Matching Based On The Array Camera

Stereo vision is a hot topic in the computer vision, after years of development, the research of stereo vision had made significant achievements and has been widely used in many fields. A complete stereo vision system consists of image acquisition, camera calibration, feature extraction, stereo matching, depth estimation and 3D reconstruction. The stereo matching is a very important step in the stereo vision system, the matching accuracy directly affect the effect of depth estimation and 3D reconstruction. While the array camera is a new product, few research had been studied for the array images. In this paper, we mainly discuss the array image stereo matching technique with the 3×3 array camera as the experimental platform. The study includes camera calibration and stereo matching to obtain the parallax images. Thanks to the multiple images of different lenses and hardware synchronization of their capturing time, the studies in this paper can effectively achieve the images stereo matching. The specific research contents are as follows:Firstly, camera calibration is introduced, we describe two camera imaging model and classic Zhang Zhengyou camera calibration method in detail. Then the MATLAB calibration toolbox is used to obtain the internal and external camera parameters of 9 lens.Secondly, the image stereo matching is highlighted, combining with the advantages of global stereo matching and local stereo matching algorithm we propose a semi-global stereo matching method of this paper, which is not sensitive to the light influence and uses mutual information to calculate the matching cost. The algorithm realizes the array images stereo matching and improves the accuracy of matching result. Then it introduces the Graph Cuts algorithm for array lens image research. The Graph Guts algorithm introduced realizes the array image stereo matching and lays a foundation for depth estimation and 3D reconstruction. The experiment sets up the mixed programming platform which is based on the MATLAB and OpenCV, the software running environment is compiled with the Visual Studio 2010 and OpenCV2.4.9.In addition to the novelties in terms of objective, theory and methodology, the algorithm is effectively verified in the 3×3 array lens. The results of the experiment are expected to provide technical support for the production of the array camera module and used in the mobile phone and ipad for practical application.