The Research And Realization On Fast Feature Point Matching Technology

Image matching is a hotspot and difficulty which people have worked on for many years, and what it aims at is that finding one or several transformation in transformation space in order to make two or more images, which come from different viewpoints or different sensors or different time, get to accordance in space. At present, image matching is used in movement analysis, examination under the water, image analysis, self-motion etc. We can realize automatic image matching using some special equipment, but the majority of them are very expensive and complicated and image matching technology provided a good solution for this problem.This thesis goes deep into the research of image feature matching. First of all, choosing reasonable features and detecting them accurately is the precondition of designing matching algorithm. To select corner detector which has high repeatability and little delocalization, the performances of many corner detectors are compared. Second, a feature point merging algorithm is proposed which is used to optimize the preliminary feature points and make them more brief and robust after detecting feature points. This algorithm can effective shorten the operation in correlation matching and the probability of error matching. At same time, this paper improved on hierarchy projection matching, made correlation matching operation from 2D to 1D, and threw away the error-matched points as soon as possible, which shorten the matching time in great deal. At last, we developed a matching software, in which we selected Harris corner detector as feature point from images, by correlation matching, computing the intensity of matching, relaxation method, Lmeds method, Epipolar constraint, we got matched points from origin images.The result of image matching disposal given in this paper shows thesoftware system is efficient and feasible in the point-matching, thus it could provide foundation for the further engineering applications.