Abstract: Particle filter approximates to the posterior probability density function with a set of weighted random sample points and realizes accurate estimation of arbitrary state model. It combines Rao-Blackwellized Particle Filter(RBPF) with Multiple Hypothesis Tracking(MHT) and separates multiple target track problem into two parts: estimation of the posterior probability distribution of data association and estimation of the single target track based on the data association. The former can be solved by Sequential Importance Resampling(SIR), and the latter can be solved by minimum mean square error estimation with Kalman filter. Experimental results show that the calculation particle count and the calculation amount can be reduced by using optimal importance distribution.

Key words: Kalman filter, sequential importance resampling, Rao-Blackwellized Particle Filter(RBPF), Multiple Hypothesis Tracking(MHT), optimal importance distribution

摘要： 粒子滤波用一组带有权值的随机采样点近似后验概率密度函数，实现对任意状态模型的精确估计。把Rao-Blackwellized粒子滤波与多假设跟踪算法相结合，将多目标跟踪问题分为2个部分，即数据关联中后验概率分布的估计和基于数据关联的单个目标跟踪估计。前者通过序列重要性重采样实现，后者使用卡尔曼滤波进行最小均方误差估计。实验结果表明，采用最优重要性分布可以减少计算所需粒子数和计算量。

关键词: 卡尔曼滤波, 序列重要性重采样, Rao-Blackwellized粒子滤波, 多假设跟踪, 最优重要性分布

CLC Number: 

• TP391