中国空间科学技术

中国空间科学技术 ›› 2021, Vol. 41 ›› Issue (5): 125-135.doi: 10.16708/j.cnki.1000-758X.2021.0075

• 论文 • 上一篇    

基于嵌入式FPGA加速ORB算法的遥感影像配准方法

赵智祎，孙婷，李峰 ，辛蕾，鲁啸天，梁亮   

  1. 1 北京信息科技大学 仪器科学与光电工程学院，北京100192
    2 中国空间技术研究院 钱学森空间技术实验室，北京100094
    3 清华大学 电子工程系，北京100084
  • 出版日期:2021-10-25 发布日期:2021-09-29

Remote sensing image registration method based on embedded FPGA accelerated ORB algorithm

ZHAO ZhiYi, SUN Ting, LI Feng, XIN Lei，LU XiaoTian, LIANG Liang   

  1. 1 School of Instrument Science and Opto Electronics Engineering, Beijing Information Science & Technology University, Beijing 100192, China
    2 Qian Xuesen Laboratory of Space Technology, China Academy of Space Technology，Beijing 100094, China
    3 Department of Electronic Engineering, Tsinghua University, Beijing 100084, China
  • Published:2021-10-25 Online:2021-09-29

PDF (PC)

337

摘要： 卫星上计算资源有限，星载嵌入式处理器处理遥感影像的配准时通常需要很长的时间。可编程逻辑门阵列（FPGA）利用其内部可编程器件可用于加速图像处理。提出了一种基于Xilinx公司的ZYNQ芯片加速ORB算法的遥感影像配准方法，可用于3000×3000像素尺寸的卫星图像配准，缩短了计算耗时，提升了ORB算法的计算能效比。利用FPGA能够实现真正的并行计算电路，实现ORB算法多支路单层流水线的并行计算结构。采用软硬件结合的方法实现架构，能够处理不同分辨率的图像，可灵活配置特征点的数量。基于设计的加速ORB配准方法，获得了较高准确率。与软件实现相比,OVS-1A遥感影像偏移精度损失低于0.05个像元;GF.4遥感影像偏移精度损失小于0.9个像元。将ORB配准算法流程应用在ZYNQ7020上，耗时减少了57.50%。

关键词: FPGA, 遥感影像, ORB, 并行计算, 嵌入式系统, 影像配准

Abstract: With the limited computing units on satellites, it generally takes a long time for the on-board embedded processor to process the remote sensing image alignment. The programmable logic gate arrays (FPGAs) can be utilized to accelerate image processing using their internal programmable components. In this paper, we proposed a remote sensing image registration method based on Xilinx's ZYNQ chip accelerated ORB algorithm. It was employed to align the remote sensing images with 3000×3000 pixels, which shortened the time consumption and improved the computational energy efficiency of ORB algorithm. The parallel computing architecture of ORB algorithm with multi-branch single-layer pipeline was realized by using FPGA that can realize a real parallel computing circuit. A combination of hardware and software was used to implement the architecture, which can handle images with different resolutions and can be configured with the number of feature points flexibly. The results show that the accelerated ORB alignment method can provide higher accuracy. Compared with software method, the offset precision loss is less than 0.05 pixel for the OVS-1A remote sensing images and is less than 0.9 pixel for the GF-4 remote sensing images. The time consumption of the ORB algorithm implemented on ZYNQ7020 is reduced by 5750%.

Key words: FPGA；remote sensing image；ORB；parallel computing；embedded system, image registration