摘要
无人机的识别与监控是目前安防领域研究的热点,现有的无人机检测方案成本过高、实现困难,存在一定的缺陷。针对此问题,文中提出一种使用最新型深度学习算法YOLOv5s的无人机光学快速识别定位追踪系统。首先通过深度学习算法实时检测是否存在无人机,并准确定位无人机的位置信息;再进一步使用KCF快速追踪算法锁定并持续追踪入侵目标;最后采取双目深度摄像头实时测算跟踪目标距离,定位位置信息后再转换输出无人机三维位置数据。所设计系统使用最新一代YOLOv5s深度学习模型,并通过改进训练模型使得其对无人机的识别达到了较高的准确率,特别是在运算速度方面,大大超过现有算法,满足高速追踪的要求。实验结果表明,相较于YOLOV3,YOLOv5s模型的准确率提高5.84%,召回率提高6.41%,推理速度提高300%。采用YOLOv5s和KCF算法相结合可稳定连续定位目标,且由于双目摄像头定位精确,全局识别速度高达80 f/s,完全具备高速追踪定位无人机的能力。
The identification and monitoring of UAV is a research hotspot in the field of security and protection at present.The existing UAV detection scheme has high cost,difficult implementation and certain defect.On this basis,an UAV optical fast identification,positioning and tracking system using the latest deep learning algorithm YOLOv5s is proposed.In the system,the deep learning algorithm is used to detect in real time whether there is a UAV and accurately locate the UAV,and then the KCF(kernelized correlation filter)fast tracking algorithm is used to lock and continuously track the intrusion target.The binocular depth camera is used to measure and calculate the tracking target distance in real time,and the three⁃dimensional location data of the UAV is converted and output after positioning the target UAV.The latest generation of YOLOv5s deep learning model is used in the desiged system,and the training model is improved to make its recognition of UAV achieve high accuracy,especially in terms of operation speed,which has greatly exceeded the existing algorithms and can meet the requirements of high⁃speed tracking.The experimental results show that,in comparison with YOLOV3,the accuracy of YOLOv5s model is increased by 5.84%,its recall rate is increased by 6.41%,and its reasoning speed is increased by 300%.The combination of YOLOv5s and KCF algorithm can stably and continuously locate the target.Since the positioning accuracy of binocular camera is high,the sytem′s global recognition speed reaches 80 f/s,which is fully capable of tracking and locating UAV at high speed.
作者
梅枫
高兴宇
邓仕超
李伟明
邹翔
MEI Feng;GAO Xingyu;DENG Shichao;LI Weiming;ZOU Xiang(Guangxi Key Laboratory of Manufacturing System and Advanced Manufacturing Technology,School of Mechanical and Electrical Engineering,Guilin University of Electronic Technology,Guilin 541004,China)
出处
《现代电子技术》
2023年第10期181-186,共6页
Modern Electronics Technique
基金
桂林电子科技大学研究生教育创新计划(2021YCXS017)
广西创新型发展专项基金项目(AA 18118002⁃3)
桂林电子科技大学研究生教育创新计划(2020YCXB01)。
