Research On Passive Binocular Vision Reconstruction Algorithm Using Label Stereo Matching With Region Correction

Binocular stereo vision has a wide range of applications in robotics,medical imaging,remote sensing,3D reconstruction,and autonomous driving.Binocular stereo matching has become a hot research subject as the core issue of stereo vision technology.Many stereo matching algorithms have been proposed and Application,but the accuracy of matching in weakly textured and occluded areas still needs to be improved.For the randomly initialized label matching algorithm,the pixel information in the corresponding matching block cannot be fully utilized when matching through affine changes or perspective transformations.This paper uses space the geometric information is used to correct the matching area to find the matching block.According to the optimization theory of Patch Match,a binocular passive vision reconstruction algorithm for labeling matching area correction is proposed based on the Local Exp algorithm.The main work is as follows:(1)First,based on the conventional graph cut(GC)algorithm,determine the correction transformation and correct the matching area based on the spatial geometric information of the label and the matching block,which can make the pixel information of the matching area of the left and right pictures sufficient use;(2)Then continuously update the selected labels,construct a global energy function through Markov random fields,and use a graph cut(GC)algorithm to find a 3D label that minimizes the global energy function,to obtain matching parallax with higher sub-pixel accuracy;(3)Finally,the left-right checking and the mean filtering are used for refinement of disparity map to improve the matching accuracy.Experiments show that this method is effective to find a good,smooth piecewise linear disparity map,which has better accuracy for dealing with low-texture region and occlusion region.