The Binocular Stereo3D Reconstruction Based On Image Segmentation And Graph Cuts

3D reconstruction has been always one of the important topics in the field of computer vision, and with the development of technology, the application of the3D reconstruction has been more and more widely, at present, it has been play an important role in the fields of robot navigation, virtual reality and medical diagnosis. In this paper, we apply the3D reconstruction based binocular stereo vision, it is one of the non-contact3D measurement, and calculate the depth information through process the two images of the same scene captured from two different perspectives, it is increasingly becoming a hot topic due to its non-contact, high efficiency and easy to realize. Stereo matching is the most important and the most critical process in the3D reconstruction, the common stereo matching method including the local method and the global method. In recent years, the global method including graph cuts and belief propagation has been researched widely due to its good performance. The paper described the theoretical principles of3D reconstruction based binocular stereo vision in detail, and focus on the two sessions of3D reconstruction process, namely the stereo matching and the3D reconstruction. The specify study including the following aspects.Firstly, in the process of stereo matching, due to the effect of a single3D reconstruction is difficult to meet the real needs, we apply the stereo matching method based on image segmentation combined with the method based on graph cuts. The method including five parts, color image segmentation, local matching, plane fitting, merge the adjacent blocks and graph cuts optimization. Firstly, employ the mean shift algorithm to divide the reference image into homogenous color blocks, the algorithm successfully remained the edge characteristics of the object. Secondly, use the adaptive support weighted self-adaptation dissimilarity measurement to estimate the initial disparity of the image. Thirdly, apply the Singular Value Decomposition method to solve the robust disparity plane fitting.In order to ensure reliable pixels sets for the segment, we filter out outliers which contain occlusion region through three rules, namely:cross checking, judging reliable region and disparity distance measurement. Fourthly, apply improved clustering algorithm to merge the neighboring segments, the geometrical relationship of adjacent planes such as parallelism and intersection is used to determine whether the two planes shall be merged. Lastly, using graph cuts to obtain the final disparity map. Secondly, in the process of3D reconstruction, obtain the3D information through calculating the disparity map obtained in the stereo matching process, reconstruction the model of the scene. Meanwhile, in order to further improve the reconstruction authenticity of the scene, we apply the texture mapping operation.Lastly, using the standard image pair and the true picture obtained under the laboratory environment to test the algorithm proposed in the paper. The experiment shows that compared with the single local matching and graph cut algorithm, the proposed algorithm can not only reduces the false matching ratio greatly, but also effectively keep the edge information of the object, and improve the result of reconstruction.