基于群结构集成势均衡多目标多伯努利滤波器的群目标跟踪算法  

Group Target Tracking Algorithm Based on Integrated Group Structure Aided by the CBMeMBer Filter

作　　者：梁潇洧 杨超群 朱鑫潮 莫磊 曹向辉[1] LIANG Xiaowei;YANG Chaoqun;ZHU Xinchao;MO Lei;CAO Xianghui(School of Automation,Southeast University,Nanjing,Jiangsu 210096,China;Key Laboratory of Measurement and Control of Complex Systems of Engineering(Southeast University),Ministry of Education,Nanjing,Jiangsu 210096,China;College of Information Science and Electronic Engineering,Zhejiang University,Hangzhou,Zhejiang 310030,China)

机构地区：[1]东南大学自动化学院,江苏南京210096 [2]东南大学复杂工程系统测量与控制教育部重点实验室,江苏南京210096 [3]浙江大学信息与电子工程学院,浙江杭州310030

出　　处：《信号处理》2024年第11期2040-2049,共10页Journal of Signal Processing

基　　金：江苏省自然科学基金(BK20230827);国家自然科学基金(62303109);东南大学至善青年学者支持计划(2242024RCB0011);中央高校基本科研业务费专项资金(MCCSE2023B02)。

摘　　要：群目标跟踪已被广泛应用于作战对抗、自动驾驶、低空防御等各个军民领域。群目标是指由多个单目标组成,并以相同的速度或方向进行移动的一组或多组目标集群。鉴于随机有限集(Random Finite Set,RFS)滤波器在处理多目标数据关联方面的优势,现有的群目标跟踪算法大多基于RFS滤波器。然而,这些滤波器在对群目标进行跟踪时大多忽略了各目标之间的关联、依赖关系等问题,为此,提出了一种基于群结构集成的势均衡多目标多伯努利(Cardinality Balanced Multi-target Multi-Bernoulli,CBMeMBer)滤波器的群目标跟踪算法。具体而言,首先通过邻接矩阵对目标群结构进行估计,即将群目标看作无向图,利用各目标之间的距离来估计群目标的邻接矩阵,进而将群目标划分为多个子群。并根据各子群中目标的运动状态,将其划分为群中心和群成员两种类别,分别建立了运动方程。而后在预测步骤中,利用估计的群结构来对目标状态进行预测。特别地,在滤波器的高斯混合实现步骤中,多个高斯分量被用来拟合相对应的各伯努利分量,但过多高斯分量的存在会降低对群目标状态估计的准确性,从而降低群结构估计的准确性,因此,在状态提取阶段,本文所提算法对高斯分量进行修剪,即对更新步骤后的每个伯努利分量中所包含的高斯分量进行筛选,只保留一个权重最大的高斯分量。最后,仿真结果表明,本文所提算法实现了对群目标的稳定跟踪,且跟踪性能优于传统的CBMeMBer滤波器。Group target tracking has become increasingly important in various fields,such as military operations,autonomous vehicles,and low-altitude defense systems.A group target consists of multiple individual targets moving together at the same speed or in the same direction.Due to the advantages of random finite sets(RFS)-based filters in handling data associations with multiple targets,most current group target tracking algorithms are based on RFS filters.However,these filters often overlook important factors such as inter-target correlations and dependencies when tracking group targets.To address these factors,a novel group target tracking algorithm based on the cardinality balanced multitarget multi-Bernoulli(CBMeMBer)filter has been proposed.This algorithm integrates the group structure into the tracking process to improve the accuracy of target tracking.Initially,the algorithm estimates the structure of the target group using an adjacency matrix,treating the group of targets as an undirected graph.By evaluating the distances between individual targets,the algorithm determines the group target’s adjacency matrix and divides the group into multiple subgroups.Based on the motion states of targets within each subgroup,the algorithm classifies them into two categories:group centers and group followers.It then establishes motion equations accordingly.In the prediction step,the estimated group structure guides the prediction of target states.In the Gaussian mixture implementation step of the proposed algorithm,multiple Gaussian components are used to model the corresponding Bernoulli components.However,an excess of Gaussian components can lead to inaccurate group structure estimations.To address this issue,the algorithm prunes Gaussian components during the state extraction step.It retains only the Gaussian component with the highest weight for each updated Bernoulli component,enhancing the accuracy of the filtering process.Finally,simulation results demonstrate the algorithm’s effectiveness at achieving stable and p

关 键 词：群目标跟踪 势均衡多目标多伯努利滤波器 邻接矩阵 高斯分量 随机有限集 

分 类 号：TP14[自动化与计算机技术—控制理论与控制工程]