面向多目标跟踪的数据关联方法研究综述  

A Survey on Data Association Approaches to Multi-Target Tracking

作　　者：李天成 谢昱昕 李固冲 范红旗[2] LI Tiancheng;XIE Yuxin;LI Guchong;FAN Hongqi(Key Laboratory of Information Fusion Technology(Ministry of Education),Northwestern Polytechnical University,Xi’an 710129,China;National Key Laboratory of ATR,National University of Defense Technology,Changsha 410073,China)

机构地区：[1]西北工业大学信息融合技术教育部重点实验室,陕西西安710129 [2]国防科技大学ATR全国重点实验室,湖南长沙410073

出　　处：《雷达科学与技术》2025年第1期10-31,共22页Radar Science and Technology

基　　金：国家自然科学基金(No.62422117,62471400,62071389);陕西省自然科学基础研究计划(No.2023JC-XJ-22);中央高校基本科研业务费专项资金。

摘　　要：目标跟踪广泛应用于要地防御、防空反导和无人驾驶等军事和民用领域。相比单目标跟踪,多目标跟踪往往涉及到未知数目的多个目标以及杂波、漏检等复杂情况,所面临的一个技术难点是数据关联,包括量测与目标航迹之间的关联以及不同传感器之间航迹关联等。本文梳理了多目标跟踪应用中数据关联主要的解决思路,首次将经典数据关联方法分为确定性数据关联和概率性数据关联,前者包括最近邻、全局最近邻和多假设跟踪,后者包括概率数据关联、联合概率数据关联和概率多假设跟踪,系统阐述和对比了这些典型算法及其扩展方法的基本原理、各自适用条件和优缺点,给出它们之间的关联性,并指出近年来智能学习、优化算法等也为数据关联问题提供了新的解决思路。特别是在视觉跟踪领域,关联跟踪与目标特征学习、场景感知密切结合,给数据关联带来了新的挑战。本文对上述各类方法和思路进行了分析和总结,并展望了未来发展趋势。Target tracking is widely used in military and civilian fields,such as attack and defense in important areas,air and missile defense,and unmanned driving.Compared with single target tracking,multi-target tracking often involves an unknown number of targets as well as clutter,missed detection and other complex situations.One of the core technical difficulties is the data association,including the association between the measurement and targets and that between target tracks from different sensors.The main solutions of the data association(DA)problem in multi-target tracking is reviewed in this paper,and the classic approaches are classified into deterministic and probabilistic DA for the first time.The former includes the nearest neighbor,global nearest neighbor and multi-hypothesis tracking,while the latter includes probabilistic DA,joint probabilistic DA and probabilistic multi-hypothesis tracking.Systematic elabora-tion and comparison of the basic principles,respective applicability,advantages and disadvantages of these classic al-gorithms and their extensions are provided,exposing their interrelationship.It is further pointed out that the intelligent learning and optimization algorithms have also provided new solutions for DA problems in recent years.Especially in the field of visual tracking,DA tracking is closely combined with target feature learning and scene awareness,which brings new challenges to DA.These various DA methods and ideas are analyzed and summarized,and the future development trend is prospected in this paper.

关 键 词：多目标跟踪 数据关联 确定性关联 概率性关联 

分 类 号：TN953.6[电子电信—信号与信息处理]