基于POS的无人机倾斜影像匹配方法  

UAV tilted images matching research based on POS

作　　者：赵政[1,2] 凌霄[3] 孙长奎[1] 李勇志[1] 

机构地区：[1]中国国土资源航空物探遥感中心,北京100083 [2]国土资源部航空地球物理与遥感地质重点实验室,北京100083 [3]武汉大学遥感信息工程学院,武汉430079

出　　处：《国土资源遥感》2016年第1期87-92,共6页Remote Sensing for Land & Resources

基　　金：国土资源部航空地球物理与遥感地质重点实验室航遥青年创新基金项目"低空无人机影像快速匹配算法研究"(编号:2013YFL11)资助

摘　　要：在充分考虑无人机倾斜影像特点的前提下,提出了一套基于POS的无人机倾斜影像匹配策略。在现有匹配方法的基础上,以全球SRTM(shuttle Radar topography mission)数据为辅助,实现了影像重叠区域预测,建立了影像间近似核线关系,剔除了匹配中的粗差点。为了解决SIFT(scale invariant feature transform)特征匹配算法计算量大,耗时长的问题,采用基于GPU(graphic processing unit)的SIFT方法,提高了SIFT的匹配效率。另外,考虑到无人机原始POS数据精度不高,因此在匹配策略中加入了逐步精化POS数据的思想。通过对多幅倾斜影像的匹配试验表明:该方法能够提供足够数量、分布均匀且点位正确的同名点。In full consideration of the characteristics of the UAV tilted images,this paper presents an UAV tilt image matching method based on POS. Compared with existing matching method,this paper firstly introduces the global SRTM data as auxiliary to forecast the images overlap area,and secondly set up an approximate relationship of epipolar line between images to exclude the gross error. As known to all,the SIFT matching algorithm has large computation and spends a lot of time,so this paper replaces it with SIFT GPU to improve operating efficiency.Specially,this paper refines the accuracy of POS step by step,because the initial accuracy of POS is not high. The experimental results of UAV tilt images from different tilt cameras shows that this method can provide a sufficient number of corresponding points which are evenly distributed and correct.

关 键 词：无人机倾斜影像 自由网平差 同名点预测 SIFT SRTM 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]