摘要
针对红外目标长时跟踪场景下跟踪置信度的可信度不足、相似目标中易出现误判所导致的跟踪错误问题,本文提出了一种基于状态判断模块的红外长时目标跟踪方法。该方法以孪生网络作为基础跟踪器,融合跟踪器输出的几何信息、判别信息、状态信息进行状态判断模块构建,包括LSTM网络堆叠适配设计、场景的自适应调节设计以及触发重检测机制设计等,实现了对红外目标跟踪状态的精准评估和长时跟踪。方法在红外长时跟踪数据集LSOTB-TIR-LT上进行了测试验证,结果表明:本文方法的F分数达到0.681,相较于经典的SiamRPN++LT算法性能提升了17.6%,特别适用于红外目标被遮挡、消失、被相似目标干扰或尺度发生显著变化时进行持续跟踪的应用场合。
In this paper,an infrared long-term target tracking method based on a state judgment module is proposed to address the problem of insufficient confidence in tracking in infrared target long-term tracking scenarios and tracking errors caused by misjudgment in similar targets.The Siamese network is employed as the base tracker.Geometric infor-mation,discriminative information,and state information output by the tracker are fused to build a state judgment mod-ule,including LSTM network stack adaptation design,scene adaptive adjustment design,and trigger re-detection mech-anism design.It achieves accurate evaluation of infrared target tracking status and long-term tracking.The proposed method is tested and validated on the infrared long-term tracking dataset LSOTB-TIR-LT,and the results show that the F-score of the proposed method reaches 0.681,representing a 17.6%performance improvement compared with the classic Siam RPN++LT algorithm.It is particularly suitable for application scenarios where infrared targets are occlu-ded,disappeared,interfered by similar targets,or undergo significant changes in scale for continuous tracking.
作者
王美怡
王向军
赵美蓉
刘峰
WANG Mei-yi;WANG Xiang-jun;ZHAO Mei-rong;LIU Feng(State Key Laboratory of Precision Measuring Technology and Instruments,Tianjin University,Tianjin 300072,China;MOEMS Education Ministry Key Laboratory,Tianjin University,Tianjin 300072,China;School of Precision Instrument and Opto-Electronics Engineering,Tianjin University,Tianjin 300072,China)
出处
《激光与红外》
北大核心
2025年第7期1065-1073,共9页
Laser & Infrared
关键词
长时跟踪
状态判断模块
孪生网络
红外目标
long-term tracking
state judgment module
Siamese network
infrared target
作者简介
王美怡(1998-),女,硕士研究生,主要研究方向为计算机视觉。E-mail:919290570@qq.com。;通讯作者:赵美蓉(1967-),女,博士,教授,博士生导师,主要从事光电测控和视觉检测方面的研究。E-mail:meirongzhao@tju.edu.cn。