提出了一种改进粒子滤波跟踪算法EMPF(expectation-maximization particle filter)。针对传统粒子滤波存在的动态模型的不确定问题，将EM算法与粒子滤波算法有效结合，将运动模型的参数作为待估量，采用EM算法来确定目标的运动模型参数，从而获得对目标状态的较准确估计。实验结果表明，当目标做复杂的转弯运动时，该算法能够显著地提高对目标运动状态的预测精度。

To deal with the uncertainties of the particle filter of the motion model, an improved particle filter EMPF (expectation-maximization particle filter) is proposed. The target states could be estimated more accurately by combining the EM and the PF algorithms, in which the parameters of the motion model are estimated and later confirmed by the EM algorithm. Thus the target states could be estimated more accurately. And the experiment results show that when the target was turning, the algorithm can improve the estimation of the target’s motions dramatically.