一种改进SURF算法的无人机影像快速配准方法  

An improved SURF algorithm-based UAV image fast registration method

作　　者：徐瑞瑞 雷添杰[2] 程结海[1] 路京选[2] 曲伟[2] XU Ruirui;LEI Tianjie;CHENG Jiehai;LU Jingxuan;QU Wei(School of Surveying and Land Information Engineering,Henan Polytechnic University,Jiaozuo 454000,Henan,China;China Institute of Water Resources and Hydropower Research,Beijing 100038,China)

机构地区：[1]河南理工大学测绘与国土信息工程学院,河南焦作454000 [2]中国水利水电科学研究院,北京100038

出　　处：《水利水电技术（中英文）》2021年第4期22-31,共10页Water Resources and Hydropower Engineering

基　　金：“十三五”国家重点研发计划项目(2017YFB0504105);国家自然科学基金项目(41601569)。

摘　　要：无人机遥感系统应用领域较为广阔,为了解决无人机影像拼接过程中存在幅宽小、数量多、重叠度不规则等问题,研究提出一种以实际应用需要为导向的无人机遥感影像快速处理方法。在SURF(Speed Up Robust Feature)算法的基础上,先采用PROSAC(Progressive Sample Consensus)算法去除大量匹配点对提高遥感影像配准的精度,然后采用GPU(Graphic Processing Unit)并行运算提高改进SURF算法的计算速度,最后再使用PROSAC几何验证实现研究区影像精准拼接。结果表明:改进SURFGPU算法的准确率比SURF算法提高了7%,像元精度达到0.4个像元;改进的SURFGPU算法用于无人机遥感影像配准的运行时间比SURF算法约少16倍,计算时间达到毫秒级。改进SURF算法具有更好的匹配精度和更快的运行速度,能满足无人机遥感影像配准速度和精度的要求,尤其适用于应急救援等实时性要求较高的应用领域。Remote sensing system fitted on UAV(Unmanned Aerial Vehicle)has been applied widely in various research fields.In order to solve the problems during the UAV(Unmanned Aerial Vehicle)image stitching,such as small coverage area,large number,irregular overlap,etc.,a practical application demand-oriented method for the fast processing of UAV remote sensing images is studied and put forward herein.On the basis of SURF(Speed-up Robust Features)algorithm,the accuracy of sensing images registration is enhanced at first by means of removing a large number of matching point pairs with PROSAC(Progressive Sample Consensus)algorithm,and then the calculating speed of the SURF algorithm is enhanced and improved with GPU(Graphic Processing Unit)parallel computing.Finally,the accurate image mosaic of the study area is realized through PROSAC geometric verification.The results show that the accuracy rate of the improved SURFGPU algorithm is increased by 7%if compared with that of the SURF algorithm and the pixel accuracy reaches 0.4 pixels,while the operation time of the SURFGPU algorithm for UAV remote sensing image registration is about 16 times less than that of the SURF algorithm and the calculating time reaches the millisecond level.The improved SURFGPU algorithm has a better matching accuracy and a faster operation speed,thus can meet the requirements of both the speed and the accuracy for the sensing image registration,which is especially suitable to be applied to those application fields with higher real-time requirement,such as emergency rescue,etc.

关 键 词：无人机 遥感影像配准 SURF算法 SURFGPU算法 并行计算 

分 类 号：P231[天文地球—摄影测量与遥感]