一种多模型贝努利粒子滤波机动目标跟踪算法

Multiple Model Bernoulli Particle Filter for Maneuvering Target Tracking

交互式多模型贝努利粒子滤波器(Interacting Multiple Model Bernoulli Particle Filter,IMMBPF)适用于杂波环境下的机动目标跟踪。但是IMMBPF将模型信息引入粒子采样过程中会导致用于逼近当前时刻真实状态与模型的粒子数减少,而且每次递推各模型间的粒子都要进行交互,存在计算量过大的缺点。为提升IMMBPF中单个采样粒子对于真实目标状态和模型逼近的有效性,该文提出一种改进的多模型贝努利粒子滤波器(Multiple Model Bernoulli Particle Filter,MMBPF)。预先选定每一个模型的粒子数,且模型间的粒子不需要进行交互,减少了计算负荷。模型概率由模型似然函数计算得到,在不改变模型的马尔科夫性质的条件下避免了小概率模型的粒子退化现象。仿真实验结果表明,所提出的MMBPF与IMMBPF相比,用较少的粒子数就可获得更优的跟踪性能。

Interacting Multiple Model Bernoulli Particle Filter（IMMBPF） is suitable for maneuvering target tracking under cluttered environment. However, when model information is introduced into particle sampling process in IMMBPF, it will lead to the number decline of particles which are applied to approaching the real state and model, and the computation load is heavy because of the interacting stage of particles in the recursion. An enhanced Multiple Model Bernoulli Particle Filter（MMBPF） is proposed to improve the effectiveness of single particle to approximate the real target state and model. The number of particles of each model is given in advance, and the posterior probability of each model is updated with the associate likelihood function, which avoids particle degeneracy without distorting the Markov property. Simulation results show that the proposed MMBPF achieves better tracking performance with fewer particles than IMMBPF.