基于SIFT特征与区域相结合的双目视觉匹配算法  

BINOCULAR VISION MATCHING ALGORITHM BASED ON COMBINATION OF SIFT FEATURES AND AREA

作　　者：黄向东[1] 姬智华[1] 

机构地区：[1]西安科技大学电气与控制工程学院,陕西西安710054

出　　处：《计算机应用与软件》2014年第4期145-148,共4页Computer Applications and Software

摘　　要：以SIFT特征对旋转、尺度缩放、亮度变化保持不变性的特点,使用双目视觉系统从两个不同的视觉角度采集立体图像对,提出基于SIFT特征匹配和改进的区域匹配相结合的匹配算法。该方法确定符合SIFT特征的边缘为可靠特征点,并确定其视差;根据视差梯度原理确定其他点的视差,最后生成稠密的视差图。实验结果表明,SIFT特征与区域相结合的匹配算法的引入,提高了特征点视差的准确性,一些弱纹理区的匹配也有所改善。SIFT feature has features of keeping invariance against the variations in rotation,size scaling and brightness,considering these features,we use binocular vision system to collect stereo image pairs from two different vision angles,and propose a matching algorithm which is based on the combination of SIFT feature matching and improved area matching. The method determines the edge conforming to SIFT features as the reliable feature point and determines its parallax; the parallaxes of other points are determined according to parallax gradient principle,and finally the dense disparity map is then generated. Experimental results show that the introduction of matching algorithm combining the SIFT features and area improves the accuracy of the parallax of feature point,the matching of some weak texture areas is also improved.

关 键 词：双目视觉 SIFT 特征匹配 区域匹配 视差梯度 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]