一种改进的SIFT图像立体匹配算法
An Improved SIFT Algorithm for Image Stereo Matching

摘要： 针对SIFT算法复杂度高、计算时间长、影响立体匹配的实时性等问题,提出了一种改进的立体视觉特征点匹配算法该算法从两个方面对SIFT算法进行改进:首先利用24维特征描述符代替128维特征描述符,以降低计算复杂度;其次在图像对匹配过程中采用改进的BBF搜索算法,通过引入最小优先级队列的限制条件和匹配精度更高的马氏距离判断两幅图像特征点的匹配性.采用经典图像和未知的室外环境下拍摄的图像对本文算法进行实验验证,结果表明,本文提出的算法每100个特征点检测时间为0.01 s,正确匹配率平均为89.65%,相对于原算法,提高了匹配的准确度,并降低了匹配时间.
Abstract: The high complexity and long computing time of SIFT (scale invariant feature transform) algorithm affect the real-time ability of stereo matching. To solve this problem, an improved feature-points matching algorithm of stereo vision was proposed. The SIFT algorithm was improved in two aspects. First, 24-dimensional feature descriptor instead of 128-dimensional feature descriptor was used to reduce computational complexity. Then the improved BBF search algorithm was used in the process of image matching, so that the feature point matching of the two images can be determined through the minimum priority queue restrictions and the Mahalanobis distance of higher matching accuracy. The classical images and images taken at unknown outdoor environment were used to validate this algorithm. Experimental results show that the proposed algorithm spends 0.01 s to detect 100 feature points, and the average correct matching rate is 89.65%. Compared with the original algorithm, it improves the matching accuracy and reduces the matching time