非T噪声条件下自适应目标跟踪方法研究  

Research on Adaptive Target Tracking Method under Non-T Noise Conditions

作　　者：张扬 马天力[1] 高嵩[1] 陈超波[1] ZHANG Yang;MA Tian-li;GAO Song;CHEN Chao-bo(College of Electronic Information Engineering,Xian Technological University,Xian Shaanxi 710021,China)

机构地区：[1]西安工业大学电子信息工程学院,陕西西安710021

出　　处：《计算机仿真》2023年第7期348-353,476,共7页Computer Simulation

基　　金：陕西省科技厅重点研发计划项目(2020GY-176);陕西省教育厅科研计划项目(20JK0674)。

摘　　要：针对目标跟踪系统中量测噪声的重尾不对称问题,提出了一种基于变分贝叶斯理论的VBST算法。算法考虑重尾不对称量测噪声条件下的卡尔曼滤波,并使之与变分贝叶斯学习结合,将量测噪声与系统状态变量一起作为参数进行联合的递归估计,通过对近似后验分布不断更新迭代,使其逼近系统的真实后验分布。仿真结果表明,在目标跟踪系统中,当量测噪声出现重尾不对称时,基于偏斜T分布的变分贝叶斯滤波算法具有较高的滤波精度和较好的鲁棒性。In order to solve the problem of heavy-tailed asymmetry of measurement noise in target tracking system,a VBST algorithm based on variational Bayesian theory is proposed.In this algorithm,the Kalman filter under the condition of heavy-tailed asymmetric measurement noise is considered and combined with variational Bayesian learning,and the measurement noise and system state variables are used as parameters for joint recursive estimation.The algorithm continuously updates and iterates the approximate posterior distribution to make it approach the real posterior distribution.The simulation results show that the variational Bayesian filtering algorithm based on skewed T distribution has higher filtering accuracy and better robustness when heavy-tailed asymmetrical distribution of measurement noise in the target tracking system.

关 键 词：分布 变分贝叶斯 卡尔曼滤波器 鲁棒滤波 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]