Abstract: To track extended targets for the linear Gaussian system, the multiple extended target Rao-Blackwellised particle filter (RBPF), which estimates the data association and multiple target states jointly, is proposed. The proposed filter applies the particle filter to estimate the data association, and employs the extended target filter based on random matrix to estimate kinematic states and shape information of extended targets. First, the framework of the multiple extended target RBPF is proposed. Then, the joint proposal distribution for the association hypothesis is defined. Furthermore, the Bayesian framework of multiple extended target tracking is implemented by the combined filter, which applies the particle filter and the extended target filter based on random matrix. In comparison with the multiple extended target filter based on JPDA algorithm and the multiple extended target filter based on probability hypothesis density, simulation results show that the multiple extended target RBPF achieves the less error of the shape estimates, and enhances the position tracking accuracy in the situation that there are spatially close extended targets.

Key words: Multiple extended target tracking, Rao-Blackwellised particle filter, data association, shape estimate

摘要： 针对线性高斯情况下的多扩展目标跟踪问题，提出了多扩展目标RB粒子滤波器(Rao-Blackwellised Particle Filter, RBPF)，对数据关联和目标状态进行联合估计，采用粒子滤波思想解决数据关联问题，采用一组适用于线性高斯的基于随机矩阵的扩展目标滤波器解决目标状态和形状的估计问题。首先提出了多扩展目标RBPF的基本框架，定义了数据关联的建议分布函数，并完成了多扩展目标滤波器贝叶斯框架下的混合实现方式。仿真实验表明，与基于联合概率数据关联(JPDA)算法的多扩展目标滤波器以及基于概率假设密度的多扩展目标滤波器相比，多扩展目标RBPF 的形状估计误差较小，并提高了目标相距较近时的位置估计性能。

关键词: 多扩展目标跟踪, Rao-Blackwellised粒子滤波, 数据关联, 形状估计