多目标跟踪中多传感器分布式控制策略
Multi-sensor distributed control strategy for multi-target tracking
摘要点击 188  全文点击 204  投稿时间:2018-09-19  修订日期:2018-12-27
查看全文  查看/发表评论  下载PDF阅读器
DOI编号  10.7641/CTA.2019.80715
  2019,36(10):1585-1598
中文关键词  多目标跟踪  多传感器控制  分布式融合  信息增益  战术重要性评估
英文关键词  Multi-target tracking  multi-sensor control  distributed fusion  information gain  tactical significance assessment
基金项目  国家自然科学基金 (61873116, 51668039, 61370037), 甘肃省科技计划项目 (18YF1GA065, 18JR3RA137) 资助.
学科分类代码  
作者单位E-mail
陈辉 西安交通大学 huich78@hotmail.com 
邓东明 兰州理工大学  
韩崇昭 西安交通大学  
中文摘要
      分布式传感器网络技术在复杂的多目标跟踪系统中发挥了极其重要的作用. 本文针对多传感器多目标跟踪中的分布式传感器控制问题, 提出了基于信息论的多传感器控制策略. 首先, 本文利用随机有限集 (Random finite set, RFS) 建模, 给出了多传感器多伯努利滤波器, 并通过一组参数化的多伯努利过程来近似多传感器多伯努利密度. 进一步的, 通过多伯努利滤波器的序贯蒙特卡罗实现, 设计采样方案对多伯努利密度进行粒子采样, 用一组带有 权值的粒子集近似多目标状态空间分布. 随后, 借助巴氏距离 (Bhattacharyya distance) 作为传感器控制的评价函数用于多个传感器的独立并行控制方案的决策制定. 作为另外一个重要内容, 本文提出了一种基于多目标战术重要 性评估的多传感器控制策略. 该控制方案旨在评估多目标战术重要性的基础上对威胁度最大的目标进行优先跟踪. 最后, 仿真实验验证了所提算法的有效性.
英文摘要
      Distributed sensor network technology plays an extremely important role in complex multi-target tracking system. Aiming at distributed sensor control problem in muiti-sensor multi-target tracking, this paper proposes some multi-sensor control strategies information-based. First, a multi-sensor multi-Bernoulli filter is presented by using Random finite set, and a multi-sensor multi-Bernoulli density is approximated by a set of parameterized multi-Bernoulli process. Further, through the sequential Monte Carlo implementation of the multi-Bernoulli filter, the sampling scheme is designed to sample the multi-Bernoulli density, and then the multi-target state space distribution is approximated by a set of weighted particles. Subsequently, the Bhattacharyya distance, as the reward function, is used for the decision making of independent and parallel multi-sensor control. As another important part, this paper proposes a multi-sensor control strategy based on multi-target tactical significance assessment, where the goal is to evaluate multi-target tactical significance and then track preferentially the maximum threat target. Finally, the simulations verify the effectiveness of the proposed algorithms.