Abstract: The SIFT（Scale Invariant Feature Transform） algorithm can not locate the object shape features accurately. To solve the problem, a stereo matching method based on both Harris corner and SIFT algorithm is proposed. The algorithm extracts image feature points using Harris operator and defines the main directions for each feature point. It calculates the 32-dimensional feature vectors of each feature point descriptor and uses Best Bin First（BBF） algorithm to calculate the Euclidean distance to match. The new algorithm reduces the SIFT algorithm’s time complexity and improves the real-time performance. Experimental results of image matching in binocular stereo vision demonstrate that the new algorithm has better performance than previously reported in the literature.

Key words: Scale Invariant Feature Transform（SIFT） feature point, Harris corners, different of Gaussian, binocular stereo vision

摘要： 针对SIFT（尺度不变特征变换）算法无法准确定位物体形状特征的问题，提出了一种结合了Harris角点和SIFT算法的立体匹配方法。在DOG尺度空间提取Harris算子作为图像的特征点并为每个特征点定义主方向，计算出特征点的32维特征向量描述子并用BBF算法检索同名特征点之间的欧式距离进行匹配。在降低SIFT算法的时间复杂度的同时提高了算法提取特征点的形状意义，在双目图像匹配实验中取得了较好的结果。

关键词: 尺度不变特征变换（SIFT）特征点, Harris角点, DOG差分尺度空间, 双目视觉