《计算机应用研究》|Application Research of Computers

无人机航拍图像实时位姿估计

Real-time pose estimation of aerial images of UAV

免费全文下载 (已被下载 次)  
获取PDF全文
作者 杨坤,黄穗斌,肖化,骆开庆
机构 1.华南师范大学 物理与电信工程学院,广州 510006;2.广州极飞科技有限公司,广州 510663
统计 摘要被查看 次,已被下载
文章编号 1001-3695(2021)10-053-3177-06
DOI 10.19734/j.issn.1001-3695.2020.12.0436
摘要 针对无人机航拍图像位姿估计采用单目视觉SLAM(simultaneous localization and mapping)时具有尺度不确定性、大场景下累积误差带来的轨迹漂移以及得到的是一个局部坐标系下的相对位姿问题,提出了一种无人机航拍图像实时位姿估计的方案。首先,实时进行视觉图像的跟踪,通过引入RTK(real-time kinematic)信息得到视觉坐标系与世界坐标系的转换关系并且解决尺度不确定性和轨迹漂移的问题,最后得到一个世界坐标系下的位姿。考虑到视觉SLAM处理的视频流会处理冗余的图像,且增加了图像的存储、拍摄和计算的压力,该方案采用处理非连续拍摄的低重叠度图像来计算位姿以避免这些问题。在真实场景下的实验结果表明,该方案的精度比当前主流的开源框架ORB-SLAM2、DSO、OpenMVG的精度更高,并且实现了整体轨迹误差的均值在10 cm以内。
关键词 视觉SLAM; 运动恢复结构; RTK; 无人机; 位姿估计
基金项目 国家自然科学基金—广东大数据科学中心项目(U1911401)
广东省科技计划项目(2015A030401086)
本文URL http://www.arocmag.com/article/01-2021-10-053.html
英文标题 Real-time pose estimation of aerial images of UAV
作者英文名 Yang Kun, Huang Suibin, Xiao Hua, Luo Kaiqing
机构英文名 1.School of Physics & Telecommunication Engineering,South China Normal University,Guangzhou 510006,China;2.Guangzhou Xaircraft Technology Co. ,Ltd. ,Guangzhou 510663,China
英文摘要 Aiming at the problems of scale uncertainty, trajectory drift caused by accumulated errors in large scenes and relative pose in a local coordinate system when monocular vision SLAM was used for pose estimation of UAV aerial images, this paper proposed a real-time pose estimation scheme for UAV aerial images. Firstly, this scheme tracked the visual image in real time, then obtained the transformation relationship between the visual coordinate system and the world coordinate system by introducing RTK information, and solved the problems of scale uncertainty and trajectory drift. Finally, it obtained a pose in the world coordinate system. Considering that the video stream processed by visual SLAM would process redundant images and increase the pressure of image storage, shooting and calculation, this scheme used the processing of discontinuous low overlap images to calculate the pose to avoid these problems. In real scene experiments, the experimental results show that the accuracy of the scheme is higher than that of the current mainstream open source frameworks ORB-SLAM2, DSO and OpenMVG, and the average of the overall trajectory error is less than 10 cm.
英文关键词 visual SLAM; SFM(structure from motion); RTK; UAV(unmanned aerial vehicle); pose estimation
参考文献 查看稿件参考文献
 
收稿日期 2020/12/10
修回日期 2021/2/1
页码 3177-3182
中图分类号 TP391.41
文献标志码 A