电子学报 ›› 2019, Vol. 47 ›› Issue (3): 521-530.DOI: 10.3969/j.issn.0372-2112.2019.03.002

• 学术论文 • 上一篇    下一篇

基于高斯混合多目标滤波器的传感器控制策略

陈辉1, 贺忠良1, 连峰2, 黎慧波2   

  1. 1. 兰州理工大学电气工程与信息工程学院, 甘肃兰州 730050;
    2. 西安交通 大学电子与信息工程学院综合自动化研究所, 陕西西安 710049
  • 收稿日期:2017-11-23 修回日期:2018-01-30 出版日期:2019-03-25
    • 通讯作者:
    • 陈辉
    • 作者简介:
    • 贺忠良 男,1993年8月出生,河北临漳县人,硕士研究生.主要研究方向为多目标跟踪中的传感器管理.E-mail:zlhe93@163.com;连峰 男,1981年10月出生,陕西岐山县人,博士,副教授,博士生导师.主要研究方向为多源信息融合、多目标跟踪.E-mail:lianfeng1981@mail.xjtu.edu.cn;黎慧波 男,1982年6月出生,湖北天门市人,博士研究生,主要研究方向为目标跟踪和传感器管理.E-mail:lihuibo@stu.xjtu.edu.cn
    • 基金资助:
    • 国家自然科学基金项目 (No.61370037,No.61873116); 甘肃省科技计划项目 (No.18YF1GA065,No.18JR3RA137)

Sensor Control Strategy Based on Gaussian Mixture Multi-target Filter

CHEN Hui1, HE Zhong-liang1, LIAN Feng2, LI Hui-bo2   

  1. 1. School of Electrical and Information Engineering, Lanzhou University of Technology, Lanzhou, Gansu 730050, China;
    2. Institute of Integrated Automation, School of Electronic and Information Engineering, Xi'an Jiaotong University, Xi'an, Shaanxi 710049, China
  • Received:2017-11-23 Revised:2018-01-30 Online:2019-03-25 Published:2019-03-25
    • Supported by:
    • National Natural Science Foundation of China (No.61370037, No.61873116); Science and Technology Project of Gansu Province (No.18YF1GA065, No.18JR3RA137)

摘要: 本文基于随机有限集的高斯混合多目标滤波器(Gaussian Mixture Multi-Target Filter,GM-MTF)提出几种传感器控制策略.首先,基于容积卡尔曼高斯混合多目标非线性滤波器,借助两个高斯分布之间的巴氏距离,推导GM-MTF的整体信息增益,并以此为基础提出相应的传感器控制策略.另外,设计高斯粒子的联合采样方法对多目标滤波器的预测高斯分量进行采样,用一组带权值的粒子去近似多目标统计特性,利用理想量测集对粒子的权值进行更新,继而研究利用Rényi散度作为评价函数,提出一种适应性更好的传感器控制策略.最后,给出基于目标势的后验期望(Posterior Expected Number of Targets,PENT)评价的高斯混合实现过程.仿真实验验证了提出算法的有效性.

关键词: 传感器控制, 多目标跟踪, 高斯混合, 有限集统计, 部分可观测马尔可夫决策过程

Abstract: This paper proposes several sensor control strategies via Gaussian mixture multi-target filter (GM-MTF) with random finite set.First,on the basis of the cubature Kalman Gaussian mixture multi-target nonlinear filter,the global information gain of the GM-MTF is deduced through the Bhattacharyya distance between the two Gaussian distributions.Then,taking advantage of this information distance,this paper proposes a corresponding sensor control strategy.Furthermore,a joint sampling method of Gaussian particle is designed to sample the predicted Gaussian component of multi-target filter.Subsequently,a set of weighted particles are used to approximate the multi-target statistical characteristic,and their weights are updated with the ideal measurement set.Next,a Rényi divergence based sensor control strategy which has better adaptability is proposed.Finally,a detailed Gaussian mixture implementation of the posterior expected number of targets (PENT) is given.Simulation results verify the effectiveness of these algorithms.

Key words: sensor control, multi-target tracking, Gaussian mixture, finite set statistics, partially observable Markov decision process

中图分类号: