基于无迹变换的多目标高斯混合粒子PHD滤波  

Gaussian Mixture Particle Probability Hypothesis Density Filter Based on Unscented Transform in Multi-target Tracking

作　　者：刘欣[1] 冯新喜[1] 孔云波[1] 王兢 

机构地区：[1]空军工程大学信息与导航学院,西安710077 [2]294969部队,上海200400

出　　处：《弹箭与制导学报》2015年第5期17-21,共5页Journal of Projectiles,Rockets,Missiles and Guidance

基　　金：陕西省自然科学基金(2011JM8023);CEMEE国家重点实验室开放基金(2014K0304B)资助

摘　　要：针对在杂波环境下,一般的高斯混合粒子PHD出现滤波精度不高、滤波发散的问题,提出了一种基于无迹变换的高斯混合粒子PHD。该算法在高斯混合粒子PHD预测的基础之上,采用无迹变换进行重要性采样,结合观测值对采样粒子进行更新,获得重要性密度函数,然后对PHD进行更新。最后,将该算法与高斯混合粒子PHD进行比较;仿真结果表明,该算法在有效提高高斯混合粒子PHD精度的同时,还能提高系统的稳定性。Considering lower estimating accuracy and filtering divergence of traditional GMP-PHD algorithm in clutter environment,a modified GMP-PHD based on unscented transform was proposed. On the basis of the GMP-PHD prediction,the algorithm applies unscented transform to importance sampling,updates the sampling particles combined with observation values to get the importance density function,and then update the GMP-PHD function. The performance of the proposed algorithm was compared with traditional GMP-PHD algorithm.The simulation results show that the proposed algorithm can promote not only the accuracy but also the stability of the system.

关 键 词：多目标跟踪 概率假设密度滤波 无迹变换 高斯混合粒子PHD 

分 类 号：TN953[电子电信—信号与信息处理]