改进的多模型粒子PHD和CPHD滤波算法

多模型粒子概率假设密度(Probability hypothesis density, PHD)滤波是一种有效的多机动目标跟踪算法, 然而当模型概率过小时,该算法存在粒子退化问题,而且它对目标数的泊松分布假设会夸大目标漏检对其势估计的影响. 针对上述问题,本文提出一种改进算法. 该算法并不是简单地对模型索引进行采样, 而是用粒子拟合目标状态的模型条件PHD强度, 在不对噪声做任何先验假设的前提下, 通过重采样实现存活粒子的输入交互,提高了滤波性能. 在此基础上, 进一步将算法在Cardinalized PHD (CPHD)的框架下加以实现,提高其目标数估计精度. 仿真实验表明,所提算法在滤波性能和目标数估计精度方面均优于传统的多模型粒子PHD算法,具有良好的工程应用前景.

Improved Multiple Model Particle PHD and CPHD Filters

The multiple model probability hypothesis density (PHD) filter is an effective algorithm for tracking multiple maneuvering targets. However, when the conditional mode probabilities have small values, there is a particle degenerate problem and the Poisson assumption for the target number distribution will lead to an exaggerating effect of missed detections on the target number estimation. To solve these problems, an improved algorithm is proposed in this paper, which approximates the model conditional probability hypothesis density of target states by particles, and makes the interaction between survival targets by resampling, without any a priori assumption of the noise. Further more, the improved algorithm is implemented in the framework of the cardinalized PHD (CPHD) filter, so as to improve the accuracy of target number estimation. The simulation results show that the improved algorithm has better performance in terms of state filtering and target number estimation, so that this algorithm will have good application prospects.