重尾非高斯定位噪声下鲁棒协同目标跟踪  

Robust cooperative target tracking under heavy-tailed non-Gaussian localization noise

作　　者：陈小波[1] 陈玲[1] 梁书荣 胡煜 CHEN Xiao-bo;CHEN Ling;LIANG Shu-rong;HU Yu(Automotive Engineering Research Institute,Jiangsu University,Zhenjiang 212013,China)

机构地区：[1]江苏大学汽车工程研究院,江苏镇江212013

出　　处：《浙江大学学报（工学版）》2022年第5期967-976,共10页Journal of Zhejiang University：Engineering Science

基　　金：国家自然科学基金资助项目(61773184);国家重点研发计划资助项目(2018YFB0105000);江苏省六大人才高峰高层次人才资助项目(JXQC-007)。

摘　　要：针对定位数据的统计特性未知且易受异常值干扰而影响协同目标跟踪性能的问题,提出一种重尾非高斯定位噪声下的鲁棒协同目标跟踪方法.该方法假设定位噪声服从多元学生t-分布,建立联合估计目标状态与定位噪声参数的贝叶斯模型.针对目标状态与噪声分布参数相互耦合而难以计算联合后验分布的问题,应用变分贝叶斯推断原理和平均场理论对后验分布进行解耦,将目标状态与定位噪声参数的联合后验分布估计问题转化为最优化问题,以交替优化的方式实现系统参数的在线递推估计.对提出的协同目标跟踪方法进行测试.仿真结果表明,当定位数据中存在未知的野值噪声时,提出的协同跟踪算法具有较好的鲁棒性.The statistical properties of the localization data are unknown and the localization data are susceptible to outlier interference,which affects the cooperative target tracking performance.Aiming at the problem,a robust cooperative target tracking method for heavy-tailed non-Gaussian localization noise was proposed.It was assumed that the localization noise followed multivariate Student’s t-distribution and a joint Bayesian estimation model of target state and localization noise parameters was constructed.To overcome the difficulty of computing joint posterior distribution owing to the coupling of target state and noise distribution parameters,the variational Bayesian inference and the mean-field theory were applied to decouple the joint posterior distribution and convert the problem of joint estimation of target state and localization noise parameters into a optimization problem.Alternative optimization was implemented to achieve recursive estimation of system parameters.The proposed cooperative target tracking method was evaluated experimentally.Simulation results showed that when unknown outliers exist in localization data,the proposed algorithm has better tracking robustness.

关 键 词：协同目标跟踪 学生t-分布 变分贝叶斯推断 野值噪声 鲁棒性 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]