Abstract: A stereo matching algorithm based on ORB（Oiented fast and Rotated BRIEF）is proposed due to that stereo matching with SURF（Speeded Up Robust Features） cannot meet the demand of real-time application. Extracting features of ORB, an array is used to record sorted order as a result of the features being sorted to decrease edge effect at the same time. Search space is reduced by the epipolar constraint, uniqueness constraint and sequence constraint. KNN（K-Nearest Neighbor）algorithm is taken as matching strategy. The matching result is obtained according to the nearest neighbor matching by computing Hamming distance of ORB descriptors. The experimental results prove that the algorithm is fast, high accuracy. Even if the image is not corrected, the accuracy is still satisfactory.

Key words: stereo matching, ORB operator, epipolar constraint, uniqueness constraint, sequence constraint

摘要： 针对利用SURF（Speeded Up Robust Features）进行立体匹配难以满足实时性需求这个问题，提出了一种基于ORB（Oriented fast and Rotated BRIEF）特征的立体匹配算法。在提取ORB特征点时，用一个数组记录该特征点因减小边缘效应而排序后的次序，结合极线约束、唯一性约束和顺序约束，来减少搜索空间，再以KNN（K-Nearest Neighbor）作为匹配策略，计算特征描述子的汉明距离，最后以最近邻匹配作为立体匹配结果。实验结果表明，该方法匹配速度快，准确度高，即使在图像未经校正的情况下，仍有较高的准确率。

关键词: 立体匹配, ORB算子, 极线约束, 唯一性约束, 顺序约束