基于深度学习的无人机多目标自动跟踪系统  

UAV multi⁃target automatic tracking system based on deep learning

作　　者：薛行贵 宋德鹏 陈思 XUE Xinggui;SONG Depeng;CHEN Si(Shanghai Crops of PAP,Shanghai 200050,China;Officers College of PAP,Chengdu 610213,China;Academy of PAP,Beijing 100012,China)

机构地区：[1]武警上海总队,上海200050 [2]武警警官学院,四川成都610213 [3]武警研究院,北京100012

出　　处：《电子设计工程》2023年第18期187-190,195,共5页Electronic Design Engineering

摘　　要：为有效提高无人机目标像素水平,实现无人机对待测目标的准确跟踪,设计了基于深度学习的无人机多目标自动跟踪系统。设计核心管控电路,调节无人机操纵平台的作用能力,再利用目标节点处理模块,将其他应用元件结合起来,完成无人机多目标自动跟踪系统的硬件设计。构建深度学习网络,通过求取目标节点特征的方式,推导方向跟踪向量的具体数值,完成基于深度学习的无人机跟踪目标检索。对比实验结果,应用该研究方法无人机目标像素在水平、竖直方向上的数值都达到了200 pix以上,验证了该系统能够准确跟踪待测目标。In order to effectively improve the pixel level of UAV target and realize the accurate tracking of UAV target to be measured,a UAV multi⁃target automatic tracking system based on deep learning is designed.Design the core control circuit,adjust the function ability of UAV control platform,and then use the target node processing module to combine other application components to complete the hardware design of UAV multi⁃target automatic tracking system.Build a deep learning network,deduce the specific value of direction tracking vector by obtaining the characteristics of target nodes,and complete UAV tracking target retrieval based on deep learning.Compared with the experimental results,the values of UAV target pixels in the horizontal and vertical directions are more than 200 pix under the application of the research method,which verifies that the system can accurately track the target to be measured.

关 键 词：深度学习 无人机 多目标跟踪 操纵平台 跟踪向量 目标像素 

分 类 号：TN0[电子电信—物理电子学]