面向应急响应的无人机图像快速自动拼接  

Fast Automatic Stitching for Images of Unmanned Aerial Vehicle in Emergency Response

作　　者：吴俣[1,2] 余涛[1] 谢东海[1,2] 

机构地区：[1]中国科学院遥感应用研究所,北京100101 [2]中国科学院大学,北京100049

出　　处：《计算机辅助设计与图形学学报》2013年第3期410-416,共7页Journal of Computer-Aided Design & Computer Graphics

基　　金：科技部国际科技合作计划资助项目(Y0I0010062);中国科学院国际合作项目(GJHZ1003);中国科学院对外重点合作项目(Y0Y00630KX)

摘　　要：根据应急响应的需求,为了快速、自动且稳定地获取遥感区域的全覆盖图像,提出一种直接适用于无人机序列图像的拼接方法.利用无人机平台上的低精度飞行控制系统与无标定的非量测相机获取序列图像,并进行快速自动配准;根据全局地面点在多视图像上的投影点集,基于单应矩阵和自由投影中心建立整体优化误差方程,并利用Levenberg-Marquardt非线性迭代求解;再根据并行几何定位拼接线的快速处理方法进行图像插值和融合,实现全流程自动化的图像输出.通过灾害应急响应的无人机遥感实验结果表明,在1h内处理完成450幅序列图像,获得定位精度小于3个像素的0.05m分辨率图像;文中方法适合无人机的序列图像配准,为应急响应提供了有效的技术支持.With the demand of emergency response, in order to obtain full cover image of remote sensing area fast, automatically and stably, this paper proposes a suitable stitching method of image sequences from unmanned aerial vehicle （UAV）. A position and orientation system with low accuracy and a non-metric camera without calibration mounted on UAV can photograph the interested area to attain the image sequences and fast automatic registration. With the global ground points and the corresponding proiection points, the whole optimum error equations are calculated using Levenberg- Marquardt algorithm for nonlinear iteration of the Homography matrices and unrestrained projective centers. Then, the parallel geometric seeking algorithm of stitching line is applied l image interpolation and image fusion are processed for whole automatic image outputting. Through the disaster emergency response experiment based on UAV remote sensing, the 0.05 m resolution full cover image with less than 3 pixel of positioning accuracy are obtained from 450 images in an hour. The fast automatic algorithm proposed in this paper is appropriate for image sequences from UAV, which can provide effective technology support for emergency response.

关 键 词：无人机遥感 序列图像配准 单应矩阵 非线性迭代 拼接线 

分 类 号：P237[天文地球—摄影测量与遥感]