改进Harris-SIFT算法在双目立体视觉中的应用

Application of improved Harris-SIFT algorithm in binocular stereo vision

针对双目立体视觉图像匹配的实时性问题,提出了一种改进Harris-SIFT算法,克服了原SIFT算法提取的特征点不是角点且耗时长等问题。该算法首先用改进Harris算子进行角点提取,然后用SIFT算子构建特征描述子,最后对提取的特征点采用欧氏距离度量点对的相似性,利用最近邻搜索策略进行特征匹配。在VC++6.0与Open CV平台上,通过实验比较了所提算法与SIFT算法的特征点提取匹配结果,证明了所提算法的有效性与实时性。

Aiming at real-time problem of Binocular stereo vision image matching, an improved Harris-SIFT algorithm is proposed, which overcomes extracted feature points are not corner points and long time-consuming issues of the original SIFT algorithm. Firstly, this algorithm uses modified Harris operator for corner point extraction, then SIFT operator is used to build feature descriptor, at last the similarity of Euclidean distance metric points of the extracted feature points and the nearest neighbor search strategy is applied for feature matching. Finally, experiments on the platform of VC ＋＋6.0 and Open CV, are given to show the efficiency of the proposed approach. It is shown that the proposed algorithm is more effective and real-time compared to the feature point extraction matching results of SIFT algorithm.