Research And Implementation Of Stereo Matching Algorithm Based On Census Transform

Stereo vision is the hot research topic in the field of computer vision,and binocular stereo vision is an important branch of stereo vision.With the development of science and technology,binocular stereo vision technology,which also continues to progress,is gradually applied to the fields such as intelligent robot navigation,medical diagnosis,vehicle navigation,virtual reconstruction,video surveillance,industrial manufacturing process control and so on.The binocular stereo vision system consists of calibration,rectification,stereo matching and triangulation,and the disparity map obtained by stereo matching directly affects the restoration result of the 3D scene,so the research of stereo matching method is significant.Firstly,this paper analyzes the traditional Census transform algorithm and gives the corresponding improved algorithm.The accuracy and the robustness is improved and the time consumption is better controlled.In order to improve the defect of the traditional Census transform algorithm,which is too dependent on the center pixel,this paper uses the mean gray value of the upper,lower,left and right directions of the center pixel to correct the gray value of the center pixel.Meanwhile,this paper proposes an improved Census transform algorithm based on the change of gradient,namely,the algorithm judges whether the center pixel is located in discontinuous region by the horizontal and vertical gradient value.If the center pixel is located in discontinuous region,this algorithm use the Census transform algorithm based on weight;otherwise,this algorithm use the traditional one.The problem of traditional Census transform algorithm,which treats indiscriminately the every pixel of the transform window,is overcame from another point of view.Then,in the stage of disparity refinement,this paper improves the weighed joint bilateral filter and combines it with left-right consistent check,slope depth compensation filter,sub-pixel interpolation and median filter.And an entire solution of disparity refinement is proposed to further improve the accuracy of disparity map.Meanwhile,this paper combines the improved Census transform algorithm,guided filter and the solution of disparity refinement,and proposes the whole stereo matching method based on Census transform.Finally,the improved Census transform algorithm,the improved weighted joint bilateral filter based on LRC and the whole algorithm are experimented respectively with the image pair provided by the Middlebury website.Results show that the proposed algorithms are not only effective,but also compared with several improved stereo matching algorithms based on Census transform,the accuracy of disparity map obtained by the proposed algorithms is relatively high.