基于变分贝叶斯的分布式融合目标跟踪  

Distributed Fusion Target Tracking Based on Variational Bayes

作　　者：胡振涛[1] 杨诗博 胡玉梅[2] 周林[1,2] 金勇[1] 杨琳琳[1] HU Zhen-tao;YANG Shi-bo;HU Yu-mei;ZHOU Lin;JIN Yong;YANG Lin-lin(School of Artificial Intelligence,Henan University,Zhengzhou,Henan 450046,China;School of Automation,Northwestern Polytechnical University,Xi'an,Shaanxi 710129,China)

机构地区：[1]河南大学人工智能学院,河南郑州450046 [2]西北工业大学自动化学院,陕西西安710129

出　　处：《电子学报》2022年第5期1058-1065,共8页Acta Electronica Sinica

基　　金：国家自然科学基金(No.61976080,No.61771006);河南省科技攻关重点项目(No.212102310298);河南省教育厅科学技术研究重点项目(No.19A413006);河南大学研究生教育创新与质量提升计划项目(No.SYL20010101)。

摘　　要：考虑目标跟踪系统中未知时变过程噪声和随机异常量测噪声对目标状态估计精度的影响,本文提出了一种基于变分贝叶斯的分布式融合目标跟踪算法.首先在分布式融合框架下,结合变分贝叶斯机理利用逆威沙特分布和学生t分布,分别对无迹卡尔曼滤波实现中的一步预测协方差矩阵和量测似然概率密度函数进行建模;进而采用平均场变分理论近似解耦噪声分布参数和状态的联合概率密度函数,并通过定点迭代方法更新状态估计和噪声分布参数;最后依据协方差交叉融合策略实现对局部状态估计融合与修正.仿真结果表明,新算法综合考虑系统非线性、过程噪声时变性和量测噪声异常性的综合影响,能够有效提高运动目标的状态估计精度,同时具有较好的自适应性和鲁棒性.Considering the influence of unknown time-varying process noise and random abnormal measurement noise in the target tracking system,a new distributed fusion target tracking algorithm based on variational Bayes is proposed.Firstly,on each local platform of the distributed fusion structure,inverse-Wishart distribution and the Student’s t distribution are chosen to model the error covariance of one-step prediction estimation and measurement likelihood in the unscented Kalman filter framework according to variational Bayes.And then,the joint probability density function of noise distribution parameters and state are approximately decoupled by mean field variational Bayesian theory,so that state estimation and noise distribution parameters can be updated by fixed-point iteration.Finally,we use the covariance intersection fusion strategy to fuse and correct all local state estimations.The simulation results show that the proposed algorithm,in which system nonlinearity,time-varying process noise and abnormal measurement noise are comprehensively considered,can effectively improve state estimation accuracy of moving targets,and has better adaptivity and robustness.

关 键 词：变分贝叶斯 学生t分布 逆威沙特分布 无迹卡尔曼滤波 协方差交叉融合 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]