The Study Of Image Matching Algorithm Based On Binocular Vision

Image Matching is one of the core problems of image processing and computervision, which is playing an important role in subsequent image analysis and imageunderstanding. In computer vision areas, as people’s cognition of the artificial intelligencemore deeply, vision system using two cameras become the current research hot spotproblems. Image matching algorithm based on binocular vision is researched in this paper.First, this paper reviews the research results and actuality about the technology ofimage matching. On this basis, the basic theory of image matching is systematicallyexpounded from definition and influence factors and process, and existing algorithm ofimage matching is classified. This paper emphasizes on the image matching algorithmbased on features.Then this thesis studies in detail that the principle of the common feature extractionalgorithms and matching search methods, and design experiments to test the algorithmperformance. The experimental results show that, compared with the Haar’s cornerdetection algorithm and SIFT, SURF feature extraction algorithm not only extract thefeature points which is invariantly to scale, translation, rotation and illumination, but alsointroduces box filters and the integral image, which is greatly reducing the time for featureextraction; In the matching method based on nearest neighbor, in comparison to the globalmethod, we speeding up the nearest point searching process through the introduction ofthe concept of KD-Tree, which is greatly improving the matching search efficiency. At thesame time, timely terminated the invalid search process through limit the search maximumnumber of leaf nodes every time inquires, and simplified the matching process by priorityqueues.Last, in binocular vision field, SURF feature extraction algorithm and improvedKD-Tree two-way searching algorithm is applied. Left and right of Cones view matchingexperiments show that, in condition of guaranteeing the accuracy of the image matching,this algorithm can satisfy the requirement of real-time computer vision.