Research On Stereo Matching Algorithm Based On Local Window In Binocular Vision

With the rapid development of computer technology,binocular vision has gradually become a hot research direction.As a key technology of binocular vision,stereo matching is widely used in intelligent robot navigation,non-contact ranging,unmanned driving and other fields.The goal of stereo matching is to obtain disparity map by matching stereo image pairs.The quality of disparity map directly affects the effect of binocular vision,so it is important to study stereo matching algorithm.The main research contents of this paper are as follows:(1)The traditional stereo matching algorithm based on local window has poor antiinterference performance,low robustness,and low matching accuracy in areas with discontinuous depth and weak texture.To solve this problem,an improved Census multi feature fusion stereo matching algorithm is proposed.Firstly,the improved Census transform based on neighborhood pixel weighting,four-way gradient information and color information are used to form a multi feature fusion matching cost to enhance reliability;Secondly,dynamic cross domain and 4-path scan line optimization are used to aggregate the cost to improve the matching accuracy of the algorithm;Finally,the disparity map is obtained by disparity calculation and multi-step disparity optimization.The algorithm not only ensures the efficiency,but also reduces the error matching rate,and the disparity map effect is improved.(2)The dynamic cross domain aggregation method has limitations,resulting in poor overall performance of the algorithm in border,repeated texture and weak texture regions.To solve this problem,based on the improved Census multi feature fusion stereo matching algorithm,the cost aggregation stage is improved,and a stereo matching algorithm based on HSV and adaptive aggregation is proposed.First,HSV color space is introduced into the dynamic cross domain,and gradient threshold and adaptive color threshold are set.Then,an adaptive dynamic cross domain is constructed to accurately establish the relationship between adjacent pixels,better use the color and spatial characteristics of pixels,and improve the matching accuracy of the algorithm.(3)In order to more intuitively present the disparity map effect of stereo matching and facilitate the objective analysis of algorithm performance,a stereo matching disparity map generation system is designed using the QT framework,which integrates the two algorithms proposed in the paper.The system includes five functional modules,namely:system login area,image selection area,algorithm selection area,operation control area,and result display area.Users can operate according to requirements,display the parallax map results corresponding to the algorithm,and then calculate and save the algorithm evaluation indicators.