Research On Binocular Stereo Matching Method Based On Edge Information

With the advancement and development of the information age,computer vision has gradually become a hot research field.Among them,binocular stereo matching plays an important role in computer vision and is widely used in computer vision.However,binocular stereo matching still faces challenges,and the accuracy of matching directly affects the effect of post-three-dimensional reconstruction.Therefore,effectively solving the matching accuracy problem in binocular stereo matching plays a key role in binocular stereo matching.The Census non-parametric transform local stereo matching algorithm is more robust to illumination in the local stereo matching algorithm,but there are still some shortcomings.For example,the central pixel in the Census transform is susceptible to noise interference,mismatching,etc.Therefore,it is easy to cause mismatching in the local stereo matching of Census nonparametric transformation,and the stereo matching is improved.In the matching cost calculation phase,the neighborhood value is used as the threshold instead of the center pixel of the window and the Gaussian color model is used to replace the traditional RGB color model,The gradient information is introduced to combine the Census matching cost,and the three methods are used to improve the matching precision of the matching cost calculation stage;In the cost aggregation phase,the minimum spanning tree algorithm is combined with the cross-scale cost aggregation method to perform cost aggregation,and the matching accuracy of the cost aggregation phase is improved.In the parallax calculation phase,the "winner is king"-WTA strategy is used to find the most Excellent difference;In the parallax optimization stage,the mismatching problem between the occlusion area and the edge depth discontinuity is mainly solved.The left and right consistency detection method is used to detect the occlusion area,and the background filling and median filtering method are used to fill and smooth the occlusion area,and the Canny edge is adopted.The operator is detected to detect the edge region,and the binary method and the guided filtering method are used to connect and smooth the edges,and the sub-pixel method is used to improve the overall matching accuracy of the improved algorithm;Finally,the improved algorithm is evaluated and analyzed by the Middlebury test platform.It can be known that the improved algorithm improves the overall matching accuracy of the Census nonparametric transform local stereo matching algorithm.