The Depth Acquisition Techniques Based On Multi-view Color Images

With the development of computer vision,the acquisition techniques of scene depth information has particular significance as one of the important subjects in three-dimensional reconstruction,and its significance is particularly important.Since the active acquisition methods of scene depth information such as the use of TOF and Kinect,are costly and seriously affected by noise.Passive depth acquisition methods represented by stereo matching are used to obtain scene depth information generally.Stereo matching has the advantages of simple principle,no coding decoding,and low requirements for scene and camera hardware.However,it is a reverse problem in itself,and many problems have not yet been solved.Meanwhile the depth information obtained by using only two viewpoints for three-dimensional reconstruction is not universal and persuasive.Therefore,this dissertation focuses on the stereo matching algorithm of multi-view,and lays a foundation for estimating the virtual viewpoint and three-dimensional reconstruction.In this paper,the principle of binocular stereo vision is introduced firstly.Then classify and explain the commonly used stereo matching algorithms,and point out the problems in stereo matching.Secondly,the stereo matching algorithm based on steady-state matching probability is focused researched.In order to solve the problem of jagged edges and parallax holes in the disparity map obtained by this algorithm,an improved stereo matching algorithm based on steady-state matching probability is proposed.This algorithm is able to solve the problem through improving the probability model and adding disparity filling criteria based on hill-climbing color segmentation.For the problem of occlusion in binocular stereo matching,a multi-view stereo matching algorithm is proposed.And the stereo matching algorithm based on steady-state matching probability is combined with the improved semi-global matching algorithm to solve the problem of partial correct parallax loss in left-right consistency detection.Finally,use the image experiments on the Middlebury test platform.The comparison and analysis of the experimental results show that the proposed algorithm can deal with the above problems well,effectively improve the matching accuracy,and has robustness.