基于高斯Cost-Reference粒子滤波器的WSN目标跟踪算法  

Target Tracking Algorithm in Wireless Sensor Networks Based on Gaussian Cost-Reference Particle Filter

作　　者：余志军[1] 魏建明[1] 刘海涛[1] 

机构地区：[1]中国科学院上海微系统与信息技术研究所,上海200050

出　　处：《系统仿真学报》2009年第2期576-580,共5页Journal of System Simulation

基　　金：国家863计划项目(2006AA01Z216)

摘　　要：提出了一种新的高斯Cost-Reference粒子滤波器算法及多传感器动态协同策略用于无线传感器网络目标跟踪问题。该算法的显著特点是:(1)鲁棒性,不需要事先对系统过程噪声和测量噪声的分布进行精确建模,具有较好的噪声自适应能力,非常适用于无线传感器网络未知的、复杂的应用场景;(2)能量有效性,该算法采用高斯分布来近似状态的后验概率分布,节点间交互时只需要传输高斯分布的均值和方差,而不需要传输所有的粒子及其权值,极大地减轻了网络通信负载,能有效延长网络的寿命。A new Gaussian Cost-Reference particle filtering algorithm and associated multi-sensor dynamic collaboration scheme were proposed for target tracking applications in wireless sensor networks （WSN）. The new collaborative tracking algorithm has two absorbing features： （1） robust. It does not assume explicit mathematical models of the noise probabilistic distributions common to generic particle filter. This feature is very important for WSN applications because of the environmental uncertainty and complexity. （2） energy-efficient. It approximates the posterior probabilistic distributions by single Gaussians. When handing over the tracking task from one node to another, only the Gaussian mean and covariance but not the raw particles and associated weights need to be propagated. The Gaussian approximation can observably reduce communication burden, thus the network lifetime is efficiently prolonged. Simulation part evaluated the proposed algorithm by the metrics of tracking precision and communication complexity.

关 键 词：目标跟踪 无线传感器网络 粒子滤波器 能量有效 鲁棒性 

分 类 号：TP301[自动化与计算机技术—计算机系统结构] TP391[自动化与计算机技术—计算机科学与技术]