现代防御技术 ›› 2018, Vol. 46 ›› Issue (4): 107-112.DOI: 10.3969/j.issn.1009-086x.2018.04.017

• 探测跟踪技术 • 上一篇    下一篇

大规模无人机遥感影像快速区域网平差

杜娟, 薛武, 赵蓓蕾   

  1. 航天工程大学,北京 101416
  • 收稿日期:2018-01-11 修回日期:2018-02-05 出版日期:2018-08-30 发布日期:2020-11-25
  • 作者简介:杜娟(1985-),女,山西太原人。博士生,研究方向空间态势感知。通信地址:101416 北京怀柔八一路一号航天工程大学 E-mail:28927066@qq.com。

Fast Block Adjustment with Large-scale UAV Remote Sensing Images

DU Juan, XUE Wu, ZHAO Bei-lei   

  1. Astronautics Engineering University,Beijing 101416,China
  • Received:2018-01-11 Revised:2018-02-05 Online:2018-08-30 Published:2020-11-25

摘要: 随着无人机遥感的快速发展和广泛应用,大规模无人机影像的区域网平差成为遥感影像处理的重要研究内容。针对大规模影像区域网平差对计算机内存开销大、时间消耗长的问题,提出了一种快速高效的解决方案。一方面对传统区域网平差解算流程进行了优化,分别采用矩阵的Schur补和预处理矩阵减小了未知数的数目和法矩阵的条件数,并利用共轭梯度的截断牛顿法进行法方程的答解,大大提高了法方程答解速度;另外一方面尽量发挥平台的计算资源优势,借助CPU和GPU设备对矩阵运算进行了加速。利用典型无人机影像进行实验,结果表明提出的解决方案可以处理上万幅无人机影像而且具有较高的效率。

关键词: 区域网平差, Schur补, 预处理矩阵, 共轭梯度, 截断牛顿法, 并行加速

Abstract: With the rapid development and widespread application of unmanned aerial vehicle (UAV) remote sensing, the block adjustment of large-scale UAV images has become an important research content of remote sensing image processing. To solve the problem of computer memory overhead and time consumption in large-scale UAV image block adjustment, a fast and efficient solution is proposed. The traditional block adjustment procedure is optimized. The matrix Schur's complement preconditioned matrix is used to reduce the number of unknowns and the matrix's condition number respectively. The conjugate gradient truncation-Newton method greatly improves the speed of the solution of the normal equations. It gives full play to the computational potential of the hardware platform and accelerates the matrix operations by parallelizing the CPU with the GPU. The experimental results show that the proposed solution can handle tens of thousands of UAV images with high efficiency.

Key words: block bundle adjustment, Schur component, preconditioned matrix, conjugate gradient, inaccurate Newton solution, parallel acceleration

中图分类号: