Dense SIFT与改进最小二乘匹配结合的倾斜航空影像匹配方法  

Oblique Aerial Image Matching Method Based on Dense SIFT and Improved Least Squares Matching Algorithm

作　　者：杨幸彬 吕京国 张丹璐 成喆 YANG Xingbin;LU Jingguo;ZHANG Danlu;CHENG Zhe(College of Surveying and Urban Spatial Information,Beijing University of Civil Engineering and Architecture,Beijing 102616,China)

机构地区：[1]北京建筑大学测绘与城市空间信息学院,北京102616

出　　处：《测绘通报》2018年第10期32-36,70,共6页Bulletin of Surveying and Mapping

基　　金：北京未来城市设计高精尖创新中心重大项目(UDC2018031321)

摘　　要：针对倾斜航空影像获取到的匹配点少、分布不均匀和匹配效率低等问题,提出了一种基于Dense SIFT与改进最小二乘匹配结合的倾斜航空影像连接点自动匹配方法。首先,利用影像间粗略定向参数对影像进行透视变形改正,消除影像间的几何变形、尺度和旋转影响;其次,对影像建立金字塔并分块,在顶层金字塔影像上逐块提取Dense SIFT采样点,并基于KD树搜索临近点后进行粗匹配,采用随机取样一致性算法剔除误匹配后获得密集、分布均匀的初始匹配点;然后,利用改进的最小二乘匹配算法对初始匹配点位置进行改正;最后,将匹配点逐层换算到下层影像,重复利用改进的最小二乘匹配算法得到最终的匹配点坐标。选取典型区域的倾斜航空影像数据进行匹配试验,结果表明,本文方法能够获取到密集、分布均匀的匹配点,且相比于SIFT、ASIFT算法匹配效率更高。An automatic tie-points matching method of oblique image is proposed based on Dense SIFT and improved least squares matching to solve the problems of less matching points,uneven distribution and low matching efficiency. Firstly,the approximate orthophoto is obtained by the perspective transformation which the geometric deformation,scale and rotation between images is eliminated. Then pyramid image is built and split into blocks. Feature points are extracted using Dense SIFT algorithm from the top pyramid image block,and matching points are searched based on the KD tree,then the random sampling consistency algorithm is used to eliminate the mismatching points. An improved least squares matching algorithm is applied to correct the location of the initial matching point. Finally,the coordinate of the matching points are obtained by improved least square matching algorithm on all level pyramid images. Images of typical region is selected to test the matching effects,the experimental results show that the method can obtain dense and uniform matching points,and the efficiency is higher than SIFT and ASIFT algorithm.

关 键 词：倾斜摄影 透视变换 DENSE SIFT KD树 最小二乘匹配 

分 类 号：P237[天文地球—摄影测量与遥感]