The Target Feature Extraction And Matching Algorithm Under Binocular Vision

Binocular stereo vision is an important branch of the field of computer vision.It is an important means to obtain target detection and depth information.It has been widely used in industrial robots and intelligent driving systems.The existing feature detection has less accurate matching information and it is difficult to calibrate the target area.Moreover,feature detection is difficult to obtain depth information in the binocular stereo vision system,and the local stereo matching algorithm can obtain accurate depth information,but it is in noise.A large number of mis-matches occur on the edges.In view of the above problems,based on the binocular vision system,the paper proposes a feature optimization algorithm for grid statistics,the minimum spanning tree stereo matching algorithm for weighted median filtering,and a target positioning method for binocular vision.The main work of the article is as follows:(1)In order to improve the number of correct matching points in the traditional feature point detection algorithm,the post processing is optimized with the method of grid statistics.In the process of post detection,the GMS grid statistical algorithm and the RANSAC algorithm are used to remove the wrong feature point information,and the problem of distinguishing the true and false features is solved.The experimental results show that the improved method outperforms the traditional methods in matching the number of feature points and the correct matching rate.(2)In order to solve the problem of mismatch rate and edge blur of minimum spanning tree stereo matching algorithm,a weighted median filter and improved matching cost algorithm are proposed.The specific method is to optimize the mismatch by using the calculation method of expanding the matching cost difference,and use the weighted median filter to preserve the edge smoothing strategy in the processing,and solve the problem of edge blurring.The experimental results show that the average mismatch rate is controlled within 7%,and the algorithm is better than the minimum spanning tree stereo matching algorithm under multiple datasets.(3)Aiming at the poor depth reduction effect of feature-based stereo matching algorithm,a method combining feature point detection algorithm and minimum spanning tree stereo matching algorithm is proposed.The location of the target is obtained by the feature detection algorithm,and the accurate parallax graph is obtained by using the minimum spanning tree stereo matching algorithm,and the sparse expansion of the feature point detection algorithm leads to the blurred problem of the parallax graph.Experimental results show that the algorithm is better than the featurebased stereo matching algorithm in depth reduction.