Research On Binocular Stereo Matching Algorithm

Binocular vision has been widely used in many fields,such as navigation,unmanned and 3D reconstruction.As a key part of binocular stereoscopic vision,binocular stereo matching technology has always been a research hotspot,in this area,binocular camera is used to shoot in different viewpoints on the same scene,then the two input images will be matched by finding the corresponding points,consequently obtaining the disparity of the corresponding points,and then the three-dimensional information in the actual scene of will be figured out.In order to improve the accuracy and efficiency of binocular stereo matching,single-sequence Census transformation and multi-scale cost aggregation techniques are proposed.However,the single-sequence Census transformation has the disadvantages of depending too much on the center point and prone to wrong match in the weak texture area.The cross-scale operation of the multi-scale cost aggregation makes the stereo matching operation inefficient.According to the above shortcomings,in this paper,we have done the following research work:(1):An algorithm called dual sequence Census transformation is proposed.We uses mid-value of gray value in a 5*5 small window to replace the gray value of the center point as the reference point,weakening the excessive dependence on the center point;And the fixed cross window is introduced as the transformed window,and dual sequence binary information is used to identify weather the points in the window and the reference point are in a same area,compare the gray value of them as well,moreover,adaptive threshold is set to determine whether the points in the window are cross-region,it will reduce the matching error of the stereo match.(2):In the step of cost aggregation,a multi-scale cost aggregation method of optimal cost per scale is proposed.The optimal cost and disparity of the images are obtained in different scales,and then the cross-scale operation is carried out.The efficiency of multi-scale cost aggregation will increase after using new method.(3):A binocular stereo matching system based on the dual-sequence Census transformation and the multi-scale cost aggregation method of optimal cost per scale is realized.The experimental results show that the error match ratio of the stereo matching system in each region is reduced,especially in the low texture region.