高斯混合扩展目标多伯努利滤波器  

A Gaussian Mixture Extended-Target Multi-Bernoulli Filter

作　　者：张光华[1,2] 连峰[1,2] 韩崇昭[1,2] 姚玲玲[1,2] 

机构地区：[1]西安交通大学智能网络与网络安全教育部重点实验室,西安710049 [2]西安交通大学电子与信息工程学院,西安710049

出　　处：《西安交通大学学报》2014年第10期9-14,共6页Journal of Xi'an Jiaotong University

基　　金：国家自然科学基金创新研究群体基金资助项目(61221063;61005026);国家"973计划"资助项目(2013CB329405)

摘　　要：针对杂波环境下多扩展目标跟踪中数据关联过程复杂的问题,提出一种可同时估计扩展目标状态和目标数的高斯混合扩展目标多伯努利(GM-ET-MBer)滤波器,该滤波器无需对测量与扩展目标进行关联。首先采用伯努利随机有限集和泊松随机有限集分别描述扩展目标的状态和观测;然后结合扩展目标状态的预测信息,推导了扩展目标状态的更新方程,并在线性高斯条件下采用高斯混合方法递推地对扩展目标的状态进行估计跟踪。与高斯混合扩展目标概率假设密度(GM-ET-PHD)滤波器相比,GM-ET-MBer滤波器有效地提高了对目标数的估计精度。仿真结果表明,GM-ET-MBer滤波器和GM-ET-PHD滤波器对目标数估计的标准偏差分别为0.267 3和0.395 3,可知所提滤波器对目标数的估计更稳定。A Gaussian mixture extended-target multi-Bernoulli （GM-ET-MBer） filter is proposed to address the complex data association. The filter can simultaneously estimate the state and the number of extended targets without data association between observations and extended targets States and observations of extended targets are modeled as a Bernoulli random finite set and a Poisson random finite set, respectively. An updated state of extended targets is derived by combining the predicted states, and then, the state of extended targets is recursively estimated in linear Gaussian models. Compared with hypothesis density （GM-ET-PHD） filter, the Gaussian mixture extended-target probability the GM-ET-MBer filter can effectively improve estimation accuracy to the number of extended targets. Simulation results show that estimation of the proposed filter to the number of targets is unbiased and the standard deviations of estimations are 0. 267 3 and 0. 395 3, respectively, for both the GM-ET-MBer filter and GM-ET-PHD filter.

关 键 词：扩展目标跟踪 随机有限集 数据关联 高斯混合 

分 类 号：TN929.5[电子电信—通信与信息系统]