航空学报 > 2024, Vol. 45 Issue (16): 329834-329834   doi: 10.7527/S1000-6893.2024.29834

无人机状态检测Kalman滤波空地目标跟踪算法

徐心宇1,2,3, 陈建1()   

  1. 1.中国农业大学 工学院,北京 100083
    2.浙江省农业智能装备与机器人重点实验室,杭州 310058
    3.浙江大学 生物系统工程与食品科学学院,杭州 310058
  • 收稿日期:2023-11-06 修回日期:2023-12-06 接受日期:2024-02-28 出版日期:2024-03-13 发布日期:2024-03-11
  • 通讯作者: 陈建 E-mail:jchen@cau.edu.cn
  • 基金资助:
    国家自然科学基金(51979275);国家重点研发计划(2022YFD2001405);浙江省农业智能装备与机器人重点实验室开放课题(2023ZJZD2306);自然资源部超大城市自然资源时空大数据分析应用重点实验室开放基金(KFKT-2022-05);深圳市科技计划项目(ZDSYS20210623091808026);虚拟现实技术与系统全国重点实验室(北京航空航天大学)开放课题基金(VRLAB2022C10);能源清洁利用国家重点实验室开放基金课题(ZJUCEU2022002);农业农村部长三角智慧农业技术重点实验室开放基金(KSAT-YRD2023005);农业农村部华南热带智慧农业技术重点实验室开放课题(HNZHNY-KFKT-202202);高等教育科学研究规划课题重点课题(23XXK0304);中国农业大学2115人才工程项目

UAV object tracking for air⁃ground targets based on status detection and Kalman filter

Xinyu XU1,2,3, Jian CHEN1()   

  1. 1.College of Engineering,China Agricultural University,Beijing 100083,China
    2.Key Laboratory of Intelligent Equipment and Robotics for Agriculture of Zhejiang Province,Hangzhou 310058,China
    3.College of Biosystems Engineering and Food Science,Zhejiang University,Hangzhou 310058,China
  • Received:2023-11-06 Revised:2023-12-06 Accepted:2024-02-28 Online:2024-03-13 Published:2024-03-11
  • Contact: Jian CHEN E-mail:jchen@cau.edu.cn
  • Supported by:
    National Natural Science Foundation of China(51979275);National Key Research and Development Program of China(2022YFD2001405);Open Fund of Key Laboratory of Intelligent Equipment and Robotics for Agriculture of Zhejiang Province(2023ZJZD2306);Key Laboratory of Spatial-temporal Big Data Analysis and Application of Natural Resources in Megacities, Ministry of Natural Resources(KFKT-2022-05);Shenzhen Science and Technology Program(ZDSYS20210623091808026);Open Project Program of State Key Laboratory of Virtual Reality Technology and Systems, Beihang University(VRLAB2022C10);Open Fund Project of State Key Laboratory of Clean Energy Utilization(ZJUCEU2022002);Open Fund of Key Laboratory of Smart Agricultural Technology (Yangtze River Delta), Ministry of Agriculture and Rural Affairs(KSAT-YRD2023005);Open Project Program of Key Laboratory of Smart Agricultural Technology in Tropical South China, Ministry of Agriculture and Rural Affairs(HNZHNY-KFKT-202202);Higher Education Scientific Research Planning Project, China Association of Higher Education(23XXK0304);2115 Talent Development Program of China Agricultural University

摘要:

针对无人机(UAV)面向空地目标进行目标跟踪过程中,发生目标离开视野、目标被遮挡、存在相似目标干扰等问题导致追踪失败的情况,提出一种基于追踪状态检测和Kalman滤波的重定位更新机制,将其与孪生全卷积网络(siamfc)跟踪器结合。以空地目标为被跟踪对象,以无人机为跟踪空地目标的跟踪者,首先,基于双峰选择、平均峰值相关能量变化率、最高响应值变化率和峰值旁瓣比变化率的检测机制检测当前的追踪状态是否异常,判断siamfc的追踪结果是否满足作为观测值的要求。其次,Kalman滤波利用目标运动的先验信息对追踪进行预测更新,当追踪状态异常时能够及时校正调整目标跟踪结果。基于LaSOT数据集完成训练,在UAV123航空数据集和自制的以无人机为目标的数据集上进行实时目标跟踪测试和对比实验。实验结果表明:该算法在UAV123上的精确率和成功率分别为66.0%和47.4%(62 帧/s),在自制的以无人机为目标的数据集上的精确率和成功率分别为72.0%和58.6%(55 帧/s),满足目标跟踪的实时性要求,且跟踪结果优于多数跟踪器。该算法在无人机为跟踪者和被跟踪对象的情况下均能完成有效目标跟踪,应对目标离开视野、部分或全部遮挡和存在相似目标干扰等挑战性场景的能力有所增强,且算法具有良好的泛化能力。

关键词: 无人机, 目标跟踪, 孪生网络, Kalman滤波, 重定位机制, 追踪状态检测

Abstract:

In the UAV’s target tracking process, tracking failure will occur when the target leaves the field of view or is occluded, and similar target interference exists. To overcome these problems, a relocation mechanism based on the tracking status detection and Kalman filter is proposed, which is combined with the fully convolutional Siamese network (siamfc) tracker. In this paper, an air-ground target is taken as the object to be tracked, and UAV is taken as the tracker to track the air-ground target. First, the detection mechanism based on the selection mechanism for double peaks, changing rate of average peak correlation energy, changing rate of maximum response value, and changing rate of peak to sidelobe ratio is used to detect whether the current tracking status is abnormal or not, and determine whether the tracking results of the siamfc satisfy the requirement of being an observed value. Secondly, the Kalman filter utilizes a priori information of the target motion to predict and update the tracking result. When the tracking status is abnormal, it can correct and adjust the tracking result in time. We use the LaSOT video dataset to train the network. Real-time object tracking tests and comparison experiments are conducted on UAV123 aerial dataset and the customized dataset with UAV as the target. The experimental results show that the tracker optimized by the mechanism has an accuracy and success rate of 66.0% and 47.4% (62 frame/s) respectively on UAV123, and 72.0% and 58.6% (55 frame/s) respectively on the customized dataset with UAV as the target. The tracking results meet the real-time requirements of UAV object tracking, and are better than the results of most trackers. The algorithm accomplishes effective target tracking with both UAVs as trackers and tracked objects, and has an enhanced ability to cope with challenging scenarios such as the target’s leaving the field of view, partial or total occlusion of the target, and similar target interference. Meanwhile, the algorithm has good generalization.

Key words: UAV, object tracking, Siamese network, Kalman filter, relocation mechanism, tracking status detection

中图分类号: