A Binocular Stereo Matching Approach Based On Dual Fusion

The human visual system cannot fully understand the complex and special scenes in the natural world,which requires the human beings to use the computer to observe and analyze all kinds of complex and special images correctly.With the rapid development of science and technology information,the application of computer vision field has appeared in every aspect of people's life.At present,binocular stereo vision system is a research hotspot in the field of computer vision.Stereo matching plays a very important role in computer vision,the essence of stereo matching is to search the corresponding points in the image pair.And using the geometric principle to recover the three-dimensional information.However,the processing of searching the corresponding points in stereo matching problem is full of challenge.The mainly task of this article is around the binocular stereo matching problem.To reduce the mismatching rate of disparity map,we present a novel stereo matching approach based on both metrics and deep learning.Key to the effectiveness of our method is the complementation of metric features and learning features that improves the accuracy of the disparity map.Firstly,we perform information fusion in terms of averaging features from absolute intensity differences(SAD),gradient absolute differences(GRAD)and deep learning model such that the cost volume is constructed.Secondly,we adopt the guided filter for aggregating the matching costs.Thirdly,the initial disparity map is obtained by the WTA algorithm.Finally,a left-right consistency check and a weighted median filter are used to refine the disparity map and remove the mismatching points.Experimental results using the Middlebury benchmark v3 demonstrates the better performance of our method in terms of the average absolute error and the root-mean-square disparity error.In order to further improve the quality of disparity map and achieve the better results,we continue to propose a stereo matching approach via dual fusion.For the matching cost computation,we fused both structure and data oriented raw matching costs.Then,we incorporate the guided filtered costs into the cross-based cost aggregation and obtain the fused aggregated costs.The fusion schemes at the two steps effectively complement each other and result in an accurate disparity map.Experiments on the Middlebury benchmark v3 demonstrate the state-of-the-art performance of our framework in terms of various metrics.Therefore,the development of the binocular stereo vision depends on the efficient solution of the stereo matching problem.Stereo matching has far-reaching significance in binocular stereo vision.