中文摘要：

为了提高双目立体视觉算法正确匹配率,本文改进了特征点提取及立体匹配算法.特征点提取部分,定义了尺度方向不变角点;在匹配计算部分,利用极线约束计算特征点的候选匹配角点,采用视差梯度约束和最大向量角准则多约束条件实现精确匹配.对本文的改进算法进行了实验室以及水池试验验证：实验室中采集小盒图片;水池试验中,拖车带动船模在试验水池航行时,采集船后部的波浪图片.实验结果显示,改进算法在水池实验图片处理中的正确匹配率比尺度不变特征变换（SIFT）算法高14%.

英文摘要：

This paper aims at improving the feature point extraction and stereo match algorithm to increase the right matching rate ol binocular stereo vision. For the feature point extraction, w e defined corners with a fixed scale and direction. To calculate the stereo match algorithm,a pole line constraint was used to match the corners. A disparity gradient constraint and maximum vector angle criterion were combined as a multi-constraint condition to determine a one- to-one exact match. A box image collected in the lab room and wave images generated by a ship model towed by a trailer in the experiment pool were used as objects to test the proposed algorithm. The re-sults show that the correct match rate ol the proposed algorithm is 14% higher than the scale invariant feature transform （SIFT） algorithm.