基于改进的PHD粒子滤波的多目标跟踪技术

有限集统计学（FISST）理论将任意时刻目标状态的集合视为多目标集值状态，而相应的传感器观测值集合被视为多目标集值观测。通过随机有限集建模并利用集合的微积分运算可推导出最优多目标贝叶斯滤波器。然而由于涉及集合微积分运算，最优多目标贝叶斯滤波器的运算量极大。概率假设密度（PHD）滤波器是最优多目标贝叶斯滤波器的一阶矩近似，可以实现在关联不确定、目标数目未知或变化情况下的多目标状态估计。相比于最优多目标跟踪技术，基于PHD滤波器的多目标跟踪技术的运算复杂度得到了有效的降低，更易于工程应用。但在密集杂波背景下PHD滤波器的粒子实现方法仍然存在运算复杂度过高的问题。本文针对密集杂波的情形，提出一种有效的杂波滤除方法，在不影响滤波性能的情况下，降低了运算复杂度，提高了滤波效率。 

Multi-target Tracking Based on Improved Particle PHD Filter

The finite set statistics theory (FISST) treats the collection of target states at any given time as a set-valued multi-target state, and the corresponding collection of sensor measurements is treated as a set-valued multi-target observation. Modeling set-valued states and set-valued observations as random finite sets (RFSs) allows the problem of dynamically estimating multiple targets to be cast in an optimal Bayesian filtering framework. This theoretically optimal approach to multiple targets tracking involves set integrals on the multi-target state space, which are computationally intractable. The PHD filter is the first order moment approximation of the optimal multi-target Bayesian filter, which can track an unknown and time-varying number of targets under association uncertainty. The computational load of the multi-target tracking method based on the PHD filter is much lower than the optimal multi-target Bayesian filtering methods, so it is more applicable to engineer application. However, the particle PHD filter is still computationally intensive in dense clutter environment. This paper proposes an approach to eliminate some of the clutter from the measurement set at any particular time. The proposed approach does not influence the estimate accuracy significantly, but it alleviates the computational complexity of the particle PHD filter and improves the efficiency of filtering.