Research On Local Matching Algorithms For Binocular Stereo Vision

As an important branch of computer vision,binocular stereo vision has received more and more attention in recent years and is widely used in military,medical,aerospace and driverless fields.Binocular stereo matching is the most important part of binocular stereo vision.Its matching accuracy and speed directly determine the application effect of binocular stereo vision system.The high-precision and real-time stereo matching algorithm has always been a hot topic in binocular stereo matching research.This thesis mainly studies the stereo matching algorithm,and learns a large number of domestic and foreign binocular stereo matching algorithms.The stereo matching algorithm mainly includes global stereo matching algorithm and local stereo matching algorithm.This thesis analyzes the advantages and disadvantages of two different methods,and selects the local stereo matching algorithm as the research object.The research focuses on the accuracy and speed of the matching algorithm,and proposes two stereo matching algorithms.The main research contents are as follows:(1)Study a large number of basic theory and matching algorithm of stereo matching,program the common cost matching and cost aggregation algorithms,and analyze the advantages and disadvantages of different methods.(2)Aiming at the problem that the existing stereo matching algorithm has more mismatched pixels in the discontinuous image depth region,an algorithm combining stereo matching with image features and joint selection is proposed.When calculating the matching cost,the algorithm adds the edge information of the image to the features of the image,and introduces the Census transform in the edge domain to the matching cost calculation.At the same time,the joint selection mechanism is introduced in the disparity optimization phase,and the edge information is also used as the boundary.The matching accuracy of the algorithm in the discontinuous area of the image is improved,and the mismatch rate of the algorithm is reduced.(3)The existing stereo matching algorithm introduces a more complicated method to solve various problems encountered in the image acquisition process,which increases the calculation amount and complexity of the algorithm,and is not suitable for some real-time requirements.Aiming at this problem,a stereo matching algorithm based on wavelet transform and image features is proposed,which uses the high-dimensional features and low-dimensional features of the image as the matching cost.The disparity optimization phase adopts sub-pixel refinement and image-guided filtering to meet the matching accuracy.Under the premise,the algorithm runs faster.