Research On 3D Reconstruction Method Based On Sparse Representation And Binocular Stereo Vision

The rapid development of computer vision has motivated a thrust of hot research in stereo vision for three-dimensional reconstruction.The three-dimensional reconstruction of the image is mainly based on the obtained two-dimensional scene image,which is analyzed and processed to get the three-dimensional information of the objects.Three-dimensional technology has been widely applied in many fields.From preserving historical heritages to commercial manufacture,and from medical diagnosis to visual navigation,the footprints of three-dimensional have been influencing a broad spectrum of the technological frontiers.The key content in three-dimensional reconstruction is how to accurately obtain the depth information of the scene.However,the depth images obtained by current computer vision system usually have the problems of disparity hole and depth discontinuity,which affect the effect of reconstruction.With regard to these problems,this paper carried out the research of three-dimensional reconstruction method based on binocular stereo vision and sparse representation.The main works can be summarized as follows:(1)This paper proposes a stereo matching method based on super pixel segmentation combining the advantages of global stereo matching method.In this mothed,the problems of holes and mismatches in disparity images obtained by traditional stereo matching methods can be solved.In order to reduce the influence of some erroneous hop pixels on other points,simple linear iterative cluster(SLIC)is introduced to divide an image into multiple regions.The disparity calculated with the matching cost is used as the initial value,and the holes of each segmentation region are fitted and optimized.Thus,a dense disparity image is obtained.The experimental results on the standard test set show that the proposed algorithm can obtain better disparity images in several test scenarios.(2)This paper proposes a depth image reconstruction method based on sparse representation.Depth image can effectively describe the spatial structure of objects.However,the depth value of the depth image obtained by the current computer vision system is discontinuous and the resolution is low,which lead to poor effect of three-dimensional reconstruction.To address these problems,this paper exploits the sparse representation method to reconstruct the depth image.In this method,a set of sparse linear combination of over-complete dictionary is used to represent the high resolution depth image.Alternating direction multiplier algorithm is utilized to solve the sparse coefficients,and the depth image reconstruction can be transformed into the sparse signal solution problem.Therefore,better three-dimensional reconstruction effects can be obtained.This paper simulates and verifies the presented method on the Middlebury standard test set.The results show that the proposed algorithm can overcome the problems of unclear edges and depth discontinuity in the traditional methods.In addition,this algorithm also performs well on real scene images.