Stetreo Matching Method Based On Mutual Information

Binocular stereo vision is an important branch of computer vision.It is mainly used to obtain the depth of the scene.It can obtain three-dimensional information from the two-dimensional image.It can be used in the fields of automatic driving and scene reconstruction.The stereo matching algorithm is from multiple frames.The depth information is obtained in the image of the dimension,that is,the technical core of the disparity map.Many scholars at home and abroad have proposed many excellent stereo matching algorithms,including local matching algorithms based on local texture information for template matching,global matching algorithms using image segmentation and minimum spanning tree,and the establishment of matching cost functions.Optimized global matching algorithm.The author analyzes the characteristics of ideal disparity map,determines the prioritization of the cost function in stereo matching,starts from the concept of entropy in information theory,analyzes the characteristics of image gray entropy,and analyzes its feasibility as a cost function of global matching algorithm.To meet the requirements of the stereo matching cost function,we continue to introduce the concept of joint entropy and mutual information,and finally choose the gray mutual information as the basis of the cost function construction.In order to enhance the constraints between neighboring pixels,let the disparity graph satisfy the smooth and complete a priori,and integrate the characteristics of the ideal binocular system,this paper adds the regularization term to the cost aggregation after constructing the cost function,and obtains the final cost function.In order to optimize the cost function,according to the characteristics of the cost function,the scan line optimization algorithm is introduced to construct the expression of the value of each pixel for different disparity values.This paper chooses to optimize the scan line from multiple directions to strengthen the neighbor.Domain point constraints,which reduce the vertical streak caused by scan line optimization.After constructing the optimization process,this paper proposes a complete algorithm flow for inputting the left and right graphs and outputting the disparity map.The algorithm selects a loop structure with a constant number of iterations.Next,the experiment is carried out using the standard binocular graph,compared with the classical local matching algorithm SAD and the global matching algorithm SPSS.The characteristics of the disparity map obtained by different methods are analyzed,and the differences of the algorithm model in the change of key parameters are analyzed.Finally,the characteristics andresults of the proposed algorithm are summarized and analyzed,and hypotheses and feasibility analysis are proposed for the possible improvement of future algorithms.