融合DSST-KCF算法的城市低空反无人机自动图像追踪系统  

An automatic image tracking system for urban low-altitude anti-UAV fusion using the DSST–KCF algorithm

作　　者：崔勇强[1] 朱彩 韦鋆霖 张安哲 白迪 CUI Yongqiang;ZHU Cai;WEI Yunlin;ZHANG Anzhe;BAI Di(School of Electronic Information Engineering,South-Central Minzu University,Wuhan 430074,China;不详)

机构地区：[1]中南民族大学电子信息工程学院,湖北武汉430074 [2]不详

出　　处：《实验技术与管理》2024年第11期36-42,共7页Experimental Technology and Management

基　　金：国家自然科学基金项目(62201621);湖北省高等学校省级教学改革研究项目(JYZD24011)。

摘　　要：面向复杂城市低空环境中的“低、慢、小”无人机的精准快速跟踪难题,提出了融合DSST-KCF算法的反无人机自动追踪系统。该系统由双光图像采集单元和显示控制单元两部分组成;双光图像采集单元采集低空图像,然后由融合DSST-KCF算法对图像进行实时处理;融合DSST-KCF跟踪算法包括位置滤波器和尺度滤波器,其中位置滤波器采用循环矩阵形成的数据集进行训练,然后预测出下一帧目标位置;尺度滤波器从目标位置中心提取出33种不同尺度下的特征进行训练,然后确定跟踪框的尺寸;最后采用DSST-KCF-云台控制策略,使无人机保持在视场中心附近实现有效跟踪。在校园环境中,利用本系统对空中无人机进行实时跟踪测试,测试结果表明,改进后算法相比原KCF算法检测成功率平均提升了31.51%,跟踪精度平均提升了43%,适用于小型运动目标的跟踪。[Objective]Current target tracking systems and algorithms for unmanned aerial vehicle(UAV)detection,designed for fixed optoelectronic equipment with a single field of view,perform well for non-moving targets but fall short in tracking accuracy and real-time requirements for small,moving UAV targets.To facilitate accurate and expeditious tracking of“low,slow,and small”UAVs in complex urban low-altitude environments,this paper proposes an automatic anti-UAV tracking system that incorporates the DSST(discriminative scale space tracker)-KCF(kernel correlation filter)algorithm.[Methods]The system comprises two principal components:The dual-light image acquisition unit and the display control unit.The gimbal camera,which is equipped with infrared and visible light cameras,handles image acquisition.The computer-based display control unit processes real-time images,calculates gimbal control parameters,and displays UAV images.These units are interconnected via gigabit Ethernet for efficient commands and image uploads and downloads.The system initially captures low-altitude images through the dual-optical unit,which are processed in real time using the fusion DSST-KCF algorithm.The proposed algorithm combines the high real-time performance of the KCF algorithm with the high performance of the DSST algorithm to achieve high real-time performance and target scale adaptation.It includes position and scale filters.The position filter uses a cyclic matrix to train the position filter,which subsequently predicts the target position in the subsequent frame.The scale filter,trained on a loop matrix data set,extracts features from the target position center at 33 scales to determine the tracking frame size.Finally,the DSST-KCF-gimbal control strategy maintains the UAV centered in the field of view,ensuring continuous and effective tracking of the target.This algorithm balances multi-scale variations while simultaneously enhancing real-time performance.[Results]The proposed system was tested for real-time UAV tracking in a campu

关 键 词：目标跟踪 相关滤波 尺度估计 自动控制 

分 类 号：TP273.2[自动化与计算机技术—检测技术与自动化装置]