基于ORB特征点改进型图像拼接技术  

作　　者：梁伟铭[1,2] 冯颖龙 朱元 吴志红 

机构地区：[1]上海汽车集团股份有限公司乘用车公司 [2]同济大学中德学院,上海200000

出　　处：《信息通信》2017年第12期13-15,共3页Information & Communications

摘　　要：目前SIFT特征点和随机采样一致性算法(RANSAC)在图像拼接领域得到了广泛的应用,但是SIFT特征点的提取以及描述符的建立需要消耗很多的资源和计算时间,相反由于ORB特征点相较而言需要更小的计算资源和时间,并且特征点个数比较多,从而在嵌入式领域当中越来越受到重视,文章在提取ORB特征点以后,首先采用传统的方法找到每一个特征点相对应的另外一张图当中最近的特征点,完成初始匹配,然后对特征点之间的匹配概率建立基于泊松分布点数学模型,从数量较大的ORB特征点点初始匹配中挑出真实的特征匹配关系,再利用随机采样一致性算法完成计算变换矩阵,从而完成图像的拼接。At present, SIFT（Scale Invariant Feature Transform） feature point and random sampling consistency （RANSAC） al- gorithmhave been widely used in the field of image mosaic,but the extraction of SIFT feature points and the establishment of descriptors need a lot of computation time. On the contrary, ORB （Oriented FAST and Rotated BRIEF）） feature pointsrequires smaller computing resources and time, after extracting the ORB feature pointsthis paper uses the traditional method to find each feature points corresponding to another map of the nearest feature points to complete the initial match, then the mathematic model based on the Poisson distribution point is established. The real feature matching relationship is picked out from the large number of ORB feature points, and the calculation matrix is completed by using the random sampling consistency algorithm to complete the image stiching.

关 键 词：ORB特征点 泊松分布模型 RANSAC算法 图像拼接 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]