Research Of The Algorithm For Three-dimensional Reconstruction Based On Stereo Vision

Three-dimensional reconstruction is very important in computer vision and computer graphics. It indicates the development of computer technology of a country and has a lot of applications in academia and industry.Capturing accurate depth information of a scene is the most critical content in three-dimensional reconstruction. In this paper, we propose a novel method that constructs a high-resolution depth map with high quality from a low-resolution depth image that is noisy and contains holes. We believe that sparse linear combination of atoms from an over-complete dictionary generates the high-resolution depth map, and the low-resolution depth map are the samples from the high-resolution depth map.The proposed method is divided into three steps. In the first step, we stereo calibrate a color camera and a depth camera so that the low-resolution depth map can be projected onto the high-resolution color image. In order to calibrate depth camera automatically, we propose a rectangular plane extraction algorithm. In the second step, we segment the scene to find the regions with continuous changing depth by combining the color texture information with low-resolution depth map. In the last step, for each region, we find an optimal high-dimensional sparse vector representing the region best in Bayesian optimization framework, and then reconstruct the whole high-resolution depth map.We implement the proposed algorithm, and apply it on Middlebury dataset and real scenes. Comprehensive quantitative comparisons show that our method outperforms existing approaches when applied on Middlebury dataset, and qualitative comparison on real scenes indicates that our algorithm performs best.