Research On An Algorithm For Binocular Stereo Vision Image Matching Based On Harris-SIFT Operator

The core of binocular vision is stereo matching.The accuracy of matching results has determined the results of 3D reconstruction in a great part.If this problem is handled well,it has a great influence on the application of binocular vision.So the matching relationship of binocular stereo vision technology has become a key topic in the machine vision.It has been widely used in autonomous navigation of the robot,spaceflight and remote control survey,industrial automation and other fields.But so far,how to promote the matching speed with the premiss of matching accuracy is also a challenging problem.The specific content of this article is as follows:Firstly,it has introduced the definition of image matching,the important framework and the conversion relationship between images are described.Based on this,it has studied the phase matching,gray matching and feature matching algorithms.After analysis and comparison of these three algorithms,it can be concluded that the feature matching algorithm is fitter to apply in the complex situations.Compared with other two methods,this method has more structural significance and shorter computation time.Especially the matching result is still very good under the noise interference.Therefore,this paper chooses a relatively better feature matching algorithm to do a detailed study.Secondly,the basic principles and experiments of FAST algorithm,Harris algorithm and SIFT algorithm are compared and analyzed.The SIFT algorithm has the best stability and robustness.Then a detailed comparison experiment is performed on the feature descriptor generation and feature point matching process of SIFT algorithm,which proved that SIFT algorithm is an algorithm that can adapt to different lighting conditions,different positions and other situations to effectively match the target.However,the SIFT algorithm has large computational complexity,high complexity,and poor real-time performance.It is difficult to meet the requirements of systems with high real-time performance.Therefore,this paper needs to improve it so that it has a better matching effect and a wider range of practicality.Finally,based on the principle and practical application of SIFT algorithm,this paper presents a stereo matching algorithm based on Harris-SIFT.In this algorithm,sub-pixel corner points extracted by multi-scale Harris algorithm are used to replace the feature points generated by the SIFT algorithm,and 32-dimensional SIFT feature vector descriptors are generated;then the BBF method of improved k-d tree is used to perform initial matching on feature points,which can efficiently find the nearest neighbors of the feature points to match the feature points,and the search strategy is changed from one-way matching to two-way matching.Finally,the improved RANSAC algorithm is used to filter the initial match.Experimental results show that the improved algorithm significantly improves the matching accuracy,reduces the execution time of the algorithm and improves the matching efficiency.On the basis of ensuring the robustness of the SIFT algorithm,it makes up for thelow accuracy and the feature points.The obvious disadvantages such as geometrical significance are more widely used in binocular vision theory.