粒子滤波器及其在目标跟踪中的应用

为了在物理条件下对目标进行精确建模，有时需要运用非线性、非高斯系统。而常规的卡尔曼滤波算法要求系统是线性高斯型的，因而不能直接用来解决非线性、非高斯问题。为了解决这一问题，人们开发出各种非线性滤波算法。一种是扩展卡尔曼算法(EKF)，它对非线性系统进行局部线性化，从而间接利用卡尔曼算法进行滤波与估算；另一种是序列蒙特卡罗算法，亦即粒子滤波器(PF)，它是最近出现的解决非线性问题的有效算法。本文简要介绍非线性跟踪的最优与次优贝叶斯算法，重点关注粒子滤波器，通过再入大气层弹道目标的例子，说明PF在目标跟踪中的应用。

Particle Filter for Target Tracking

For many applications, it is becoming important to involve some elements with nonlinearity and non-Gaussionity in order to model accurately the underlying dynamics of a physical system. Moreover, in this case,it is usually difficult to directly utilize Kalman filter(KF),because KF is based on condition of linear-Gaussian system. For solving these problems, several variants of nonlinear/non-Gaussian filters are developed. One is extended Kalman filter(EKF), it makes a local linearization to a nonlinear system, so KF can be utilized indirectly to filter and estimate.The other is sequential Monte Carlo method based on point mass,which is called particle filter(PF). It is an efficient method dealing with nonlinear/non-Gaussian problems. In this paper, we review both optimum and suboptimum Bayesian algorithms for nonlinear tracking problems, with focus on PF. And the performances of the PF are discussed and compared with the EKF through tracking a re-entry ballistic object.