采用GPU加速的无人机光通信系统对准技术  

Alignment and Tracking Technology of Light and UAV Optical Communication System using GPU Acceleration

作　　者：徐声振 田明[1] 倪小龙[1] 于信[1] 白素平[1] XU Shengzhen;TIAN Ming;NI Xiaolong;YU Xin;BAI Suping(School of Opto-Electronic Engineering,Changchun University of Science and Technology,Changchun 130022)

机构地区：[1]长春理工大学光电工程学院,长春130022

出　　处：《长春理工大学学报（自然科学版）》2022年第6期29-36,共8页Journal of Changchun University of Science and Technology(Natural Science Edition)

基　　金：中国博士后科学基金项目(2017M621179);吉林省科技发展计划项目(20170521001HJ,20200401054GX);长春理工大学青年基金项目(XQNJJ-2019-01,XJJLG-2018-20)。

摘　　要：介绍基于图像预处理的灰度加权算法实现跟踪激光光斑质心的基本理论和方法;采用串行程序并行化、改进存取模式、算法归约化的方法对灰度加权算法进行并行优化;设计无人机跟踪实验对基于本方法所实现的光斑跟踪的实时性和准确性进行验证。实验结果表明:对于相机采集到的1280×1024像素的光斑图片,Jetson Xavier NX在GPU-CPU异构体系结构下完成灰度加权算法,所用时间约3.4 ms,相较于中央处理器所用时间10.9 ms提升了3~4倍;考虑中央处理器读取像素及闭环控制的时间,采用GPU并行加速的光斑质心跟踪帧率可达31 FPS,脱靶量小于0.1 mrad,满足了光通信系统的要求。Introduce the basic theories and methods of gray-scale weighting algorithm based on image preprocessing to track the centroid of laser spot;use serial program parallelization,improved access mode,and algorithm reduction to optimize the gray-scale weighting algorithm in parallel;no design The human-machine tracking experiment verifies the real-time and accuracy of spot tracking based on this method.The experimental results show that for the 1280×1024 pixel spot image collected by the camera,Jetson Xavier NX takes about 3.4 ms to complete the gray-scale weighting algorithm under the GPU-CPU heterogeneous architecture,which is compared with the 10.9 ms time taken by the central processing unit.It is increased by 3~4 times;considering the time of the central processing unit to read pixels and closed-loop control,the frame rate of the beam centroid tracking with GPU parallel acceleration can reach 31 FPS,and the miss distance is less than 0.1 mrad,which meets the requirements of the optical communication system.

关 键 词：灰度加权算法 光斑质心 并行优化 图像处理 嵌入式平台 

分 类 号：O439[机械工程—光学工程]