EM框架下实现被动测距的状态估计和参数学习

在红外成像跟踪系统中,通常仅能测量目标的角度信息,不能直接测量目标与观测站间的距离。研究了基于红外成像系统的被动测距技术,首先利用状态空间模型的分析方法建立被动测距的状态估计和参数学习的混合估计模型,然后介绍EM的基本原理和参数的最大似然估计。EM算法的E步利用粒子滤波和粒子平滑器来完成,实现被动测距的状态估计;M步利用梯度搜索的方法来求解参数。被动测距是一个带有未知参数的非线性系统的状态估计,文中利用状态估计与参数学习的状态空间模型来描述,并利用EM法来求解,为被动测距的求解提供了一条新的途径。模拟实验表明,基于粒子滤波和梯度搜索的EM方法能同时完成被动测距的状态估计和参数学习。

State estimation and parameter study for passive ranging using EM

Only angle can be measured directly in the infrared search and track system,but the distance between target and observer can not.Passive ranging based on infrared imaging system was studied for target position.State space model of state estimation and parameter study of passive ranging was presented in this paper.Expectation-maximization(EM) algorithm was used for parameter maximization-likelihood(ML) estimation.E-step in the EM was finished by particle filter and particle smoother,which could be used to calculate state estimation simultaneously.M-step of EM was completed with gradient search methods.Passive ranging was a nonlinear state estimation with unknown parameter.The state space model and EM method were introduced.This is a new way for the passive ranging.Simulation experiment indicates that this algorithm is effective for state estimation and parameter study of passive ranging.