Research On Feature Point Matching Algorithm Base On Binocular Vision

Binocular stereo vision technology was developed by two cameras at different angles to capture images of the same object, and matching the images to obtain three-dimensional geometric information of the object. In recent years, this technology has been continuously studied, in various fields of computer vision, as motion estimation, pattern recognition, scene analysis has and so on been widely used. The core research section of binocular vision system is images stereo matching technology, the results of matching is good or bad, depending on the stereo matching algorithm whether is both high speed and accuracy. However, due to the image matching process is affected by a variety of interfering factors, it is not satisfy the real-time performance and practicability of image processing. Stereo matching technique is the most difficult steps in binocular vision research, and constantly in the development and improvement.Base on the principle of binocular stereo vision technology, this paper study the principle of stereo matching technology and related technologies in detail, then makes a deep research on the theory of feature point extraction algorithms and the theory of SIFT stereo matching algorithm. Firstly, against the issues of Harris algorithm for image processing for extracting feature points less real-time, the calculation is larger, and sensitive to noises, proposes a Harris corner detection algorithm combine with pixels’ gray level difference. In this algorithm, the point is detected takes contradistinction from adjacent 16 pixels on the circumference of a circle that radius equal to 3, calculate the number of non-similar pixels and judged as a candidate corner and then calculate the Harris corner response function to extract corner, finally, combine with the idea of SUSAN algorithm to removes false corners, so the Harris corner detection algorithm’s speed and accuracy has been improved. The last, combines with feature descriptor and stereo matching of SIFT algorithm, one modified SIFT algorithm is proposed based on Harris corner point extraction. Firstly, the algorithm extracts the image’s corner points with improved Harris corner detection algorithm, then it generates the feature descriptors by SIFT characterization Operators, at last, it matches to feature points by using the similarities of descriptors and get matching results.Experimental results show that the algorithm is more timely and accurate than SIFT algorithm.