基于深度强化学习的多目标跟踪技术研究  

Research on Multi-target Tracking Technology Based on Deep Reinforcement Learning

作　　者：杨麒霖 刘俊[1] 管坚 莫倩倩 陈华杰[1] 谷雨[1] 石义芳 YANG Qilin;LIU Jun;GUAN Jian;MO Qianqian;CHEN Huajie;GU Yu;SHI Yifang(Key Laboratory of Fundamental Science on Communication Information Transmission and Fusion Technology,Hangzhou Dianzi University,Hangzhou 310018,China)

机构地区：[1]杭州电子科技大学通信信息传输与融合技术国防重点学科实验室,浙江杭州310018

出　　处：《无线电通信技术》2024年第1期187-192,共6页Radio Communications Technology

基　　金：浙江省自然科学基金(LZ23F030002)。

摘　　要：在随机有限集多目标跟踪过程中,由于跟踪问题的复杂性,会耗费大量的计算成本,特别是在目标和杂波密集的复杂情况中,计算成本呈指数增长。随机有限集中通常采用的分配算法——例如Murty算法的时间复杂度为滤波器生成的代价矩阵规模的三次方。为了减少跟踪耗时,结合组合优化的思想,将代价矩阵重定义为二分图,采用了一种基于深度强化学习的二分图匹配算法,取代传统随机有限集中的分配算法,并通过仿真实验验证了所提方法的可行性。实验表明,所提方法在保证跟踪性能的前提下减少了跟踪耗时,提升了跟踪实时性。In the process of multi-target tracking in random finite sets,due to the complexity of the tracking problem,it can consume a lot of calculation costs.Especially in complex situations where targets and clutters are dense,the calculation cost increases exponentially.The time complexity of the assignment algorithm commonly used in random finite sets,such as the Murty algorithm,is the cubic of the size of the cost matrix generated by the filter.To reduce tracking time,this paper integrates the concept of combinatorial optimization,reformulates the cost matrix as a bipartite graph,and adopts an online bipartite graph matching algorithm based on deep reinforcement learning to replace the traditional allocation algorithm in random finite sets.The feasibility of the method is confirmed through simulation experiments.Experiments demonstrate that this method reduces tracking time while maintaining tracking performance and enhancing the real-time efficiency of tracking.

关 键 词：随机有限集 强化学习 组合优化 

分 类 号：TN953[电子电信—信号与信息处理]