标签多伯努利机动目标跟踪与分类算法  

A Tracking and Classification algorithm for Maneuvering Targets with Labeled Multi-Bernoulli

作　　者：彭华甫[1,2] 黄高明 田威[1] 邱昊 满欣[1] PENG Huafu;HUANG Gaoming;TIAN Wei;QIU Hao;MAN Xin(College of Electronic Engineering,Naval University of Engineering,Wuhan 430033,China;Unit 92773 of the PLA,Wenzhou,Zhejiang 325807,China)

机构地区：[1]海军工程大学电子工程学院 [2]解放军92773部队

出　　处：《西安交通大学学报》2019年第2期157-162,178,共7页Journal of Xi'an Jiaotong University

基　　金：国家自然科学基金资助项目(61501484);中国博士后科学基金资助项目(2017M613370)

摘　　要：针对密集杂波下现有的多机动目标跟踪算法性能衰减严重的问题,提出了一种标签多伯努利目标跟踪与分类算法。首先,引入类别信息对目标状态进行扩维;然后利用类别属性对目标机动模型转移密度进行修正,并推导新的状态转移密度函数,抑制了错误机动模型对目标状态预测的影响;同时,建立目标位置与属性的联合量测似然函数,增大了目标与杂波的区分度,从而增强杂波抑制能力;最后,基于多模型标签多伯努利滤波器框架推导了新的预测、更新方程。仿真实验结果表明:所提算法在高杂波环境下仍能对多机动目标进行有效跟踪,其目标数估计误差及最优子模式分配距离分别约为多模型概率假设密度联合检测、跟踪、分类滤波器的1/2和1/4,为多模型势平衡多伯努利联合检测、跟踪、分类滤波器的3/4和1/2。A tracking and classification algorithm for targets with labeled multi-Bernoulli(LMB)is proposed to solve the problem that the performance of existing tracking algorithms for multiple maneuvering targets seriously degrades under dense clutter environment.Firstly,target state is extended by introducing category information.Secondly,the transfer density of the target maneuver model is modified using attributes of the target class.The impact of a wrong maneuver model on the target state prediction is suppressed,and the new state transition density function is derived.In addition,a joint likelihood function for the attributes and position of the target is established to increase the discrimination between target and clutters,and the clutter suppression ability is enhanced.Finally,an improved prediction and update equation are derived based on the multiple model labeled multi-Bernoulli filter framework.Simulation results show that the proposed method still trackes multiple maneuvering targets even in a high clutter environment.The estimation error for the number of targets and the optimal subpattern assignment distance of the proposed method are 1/2 and 1/4 respectively,of those from the multiple model probability hypothesis density joint detection,tracking and classification(JDTC)filter,and 3/4 and 1/2,ksptctialy,of those from the multiple model cardinality balanced multi-target multi-Bernoulli JDTC filter.

关 键 词：多目标跟踪 机动目标 分类 标签多伯努利 目标类别 

分 类 号：TN391[电子电信—物理电子学]