一种改进的CPHD多目标跟踪算法

CPHD(Cardinalized Probability Hypothesis Density)滤波是一种杂波环境下可变目标数的多目标跟踪算法,该文针对算法中存在的目标漏检问题提出一种改进算法,该算法在高斯混合框架下实现贝叶斯递归,通过对各个高斯分量进行标记,对目标进行航迹关联,在此基础上对修剪合并后各个高斯分量的权值进行两次分配。首先对超过检测门限的高斯分量权值进行分配,有效解决了目标漏检问题,然后基于一个目标只可能产生一个观测的事实进行第2次分配,改善了目标发生交叉时的算法性能。实验结果表明,所提方法在多目标状态估计和航迹维持方面均优于普通的CPHD算法。

Improved CPHD Filter for Multitarget Tracking

The Cardinalized Probability Hypothesis Density (CPHD) fiter is a recursive Bayesian algorithm for estimating multiple target states with varying target number in clutter. Due to the fact that there is a missed detection problem in the CPHD filter, an improved algorithm is proposed, which provides a closed-form solution under Gaussian mixture assumptions. Firstly, the estimate to track association is made by labeling each Gaussian component, and then the weights of Gaussian components having been pruned and merged are reassigned twice. At first, the Gaussian components’ weights exceeding the detection threshold are reassigned, which can solve the missed detection issue effectively, and then the second distribution is made based on the fact that a target can only have one measurement, which improves the performance when the targets cross each other. Simulation results show that the improved CPHD filter has advantages over the ordinary one in both the aspects of multi-target state estimation and track maintenance.