Research On Image Matching Algorithm Based On Binocular Stereo Vision

Image matching is a fundamental problem in computer vision, remote sensing, medical science and many other fields. In computer vision, image matching is an essential key step in achieving change detection, image correction, three-dimensional reconstruction and other applications. There are two general methods in Image matching, including gray-based and feature-based matching method. The former has a high precision, but a large computation. The latter which has a relatively small computation and better noise immunity is easy to implement real-time match.This paper studies the basic theory and key technology of image feature extraction and matching. In the feature extraction algorithm, two classical algorithms Harris corner detection algorithm and the SIFT algorithm is analyzed experimentally and evaluated after simulation. An improved Harris-SIFT algorithm is proposed aiming at the problem that feature points extracted by SIFT operator and is too much and few corners among them, firstly the modified Harris operator is used to,extracted corner, then the SIFT operator is used to built feature descriptor, which overcome the issue that feature point extracted by original SIFT algorithm isn’t the corner and is time-consuming. The matching algorithm usually shows clustering form clusters because of the actual data. In order to improve the efficiency of image matching, the two-way matching strategy based on nearest neighbor Kd-tree algorithm is surveyed to measure the similarity between the two images by calculating the Euclidean distance between feature vectors and finally determine the matching feature points.The algorithm can meet the requirements of real-time and accuracy of matching points of binocular vision algorithms. Its matching process can be:improved Harris corner is used to detect operator and extract image feature points, on the other hand SIFT descriptors is used to describe parameters for image feature points and generate feature vector. The matching search method based on improved nearest neighbor Kd-tree search is applied to search, and a reverse search followed. If the ratio between the distance of the nearest and next nearest match is less than the distance threshold value, the points in main image feature and secondary image are matching points. Finally, consistency random sampling method is used to eliminate pseudo matching points from all the matching points. On VC++6.0and Open CV platform, experiment is performed between SIFT feature matching algorithm and proposed matching algorithm aiming at the index of matching correct rate. Experiment results show that the improved matching algorithm get a higher matching correct rate and better stability. The outline of the image could be shown better and the algorithm is suitable for higher requirement of precision and real-time case.