Research On Local Stereo Matching Technique Based On Binocular Vision Ranging

As one of the important technologies in machine vision,binocular stereo vision ranging has played an important role in many fields such as intelligent machinery,three-dimensional reconstruction and military active ranging.Relevant scholars have carried out research work on binocular ranging for many years,and have achieved fruitful results in local stereo matching technology;however,the problem of significantly lower accuracy still exist when such technologies deal with areas with low texture,occlusion,complex contour images and uneven lighting.This article analyzes and studies the above problems and does the following work:First,the research on the development status and basic theoretical knowledge of binocular stereo vision at home and abroad was carried out.At the same time,the advantages and disadvantages of several local stereo matching technologies commonly used in the experiment were compared and analyzed,which provided basic knowledge reserves and theoretical basis for the improvement of later algorithms..Secondly,for the problem that the adaptive window algorithm in local stereo matching when matching images,the accuracy is susceptible to uneven illumination and the shape of the window is difficult to effectively describe the boundary of the image to be matched,a heterogeneous adaptive window local stereo matching algorithm is proposed.Before calculating the matching cost,the algorithm guides filtering preprocessing on the matched image to smooth the image to maintain the boundary.In view of the shortcomings of the traditional Census Transform that is easily affected by the fluctuation of the central pixel,the three-dimensional pixel information,on the basis of the Census Transform,is proposed to calculate the matching cost of the information,combining the inner non-center pixel difference and the center pixel difference.Finally,in order to fit the image boundary and contour better than what the traditional algorithm does,and thus improve the matching accuracy,an aggregation contrived by a special-shaped window resulting from the double helix path method is also proposed.This method adaptively determines the shape size along the two spiral search paths in the area around the central pixel at the same time,forming a more efficient and variable matching window than the traditional algorithm,thereby obtaining a high-precision parallax map.Thirdly,to solve the problem of large parallax error in low-texture areas of the image proposed by the former algorithm,a stereo matching algorithm for low-texture areas of image based on reconstruction plane is proposed.Based on the traditional superpixel segmentation algorithm,it improves to obtain a predictable superpixel segmentation number k and can merge "homogeneous" superpixels to form an improved superpixel segmentation algorithm for low-texture regions,and uses a fast straight line segment extraction algorithm to identify Low-texture area contour;at the same time,this paper uses the proposed "anchor point method" to filter out the true boundary pixel points of the low-texture area,and combines with the disparity value at the boundary in the previous chapter to construct a number of three-dimensional space points,and then fits it as a three-dimensional low-texture plane in the world coordinate system;finally,the plane equation is used to recalculate the internal parallax of the refined lowtexture area.After all the above algorithms are processed,a local stereo matching algorithm with relatively good performance in uneven illumination,complex image contours and low texture areas can be obtained.Finally,based on the Matlab2018 a software of the Windows system,combined with the improved local stereo matching technology,a binocular stereo vision ranging software module was developed.Experiments have shown that most of its ranging errors are less than 5%,which can meet the actual use needs.