嵌入式容积粒子PHD多目标跟踪算法  

Imbedded Cubature Particle PHD Filter Multi-target Tracking Algorithm

作　　者：熊志刚[1] 黄树彩[1] 赵炜[1] 苑智玮 

机构地区：[1]空军工程大学防空反导学院,陕西西安710051

出　　处：《信号处理》2016年第6期676-683,共8页Journal of Signal Processing

基　　金：陕西省自然科学基础研究计划资助项目(2012JM8020);航空科学基金(20130196004)

摘　　要：针对基于概率假设密度算法(Probability Hypothesis Density,PHD)的非线性多目标跟踪估计精度不高、滤波发散、实时性差等问题,提出一种嵌入式容积粒子PHD算法(Imbedded Cubature Particle PHD,ICP-PHD)。新的算法在采样阶段引入Halton点集,并基于三阶嵌入式容积准则产生有限的积分点,对每个采样粒子进行滤波,来拟合重要密度函数。由于Halton点集得到的粒子分布更加均匀,故而ICP-PHD算法能够避免"粒子聚集"的现象。另外,由于三阶嵌入式容积准则的积分点少、精度高,因此ICP-PHD算法能更好的协调时间与精度之间的矛盾。仿真结果表明ICP-PHD能对多目标进行有效跟踪,相比高斯厄米特粒子PHD算法(Gauss Hermite Particle PHD,GHP-PHD)具有实时性强的优势,在目标数目和状态估计上比容积粒子PHD算法(Cubature Particle PHD,CP-PHD)精度更高。Considering the low accuracy, filter divergence and poor timeliness of nonlinear multi-target tracking based on probability hypothesis density （PHD） , a new filter named imbedded cubature particle PHD （ICP-PHD） is proposed. ICP- PHD implements particle sampling with Halton points sets, and generates infinite integral points based on the third-degree imbedded cubature rule to perform particle filtering for the purpose of matching the important density function. As a result of the well-distributed particles obtained with Hahon sets, ICP-PHD can avoid the phenomenon of particle aggregation. Be- sides, ICP-PHD can deal with the contradictions between time and accuracy well because of the few integral points and high accuracy. Simulation was made and it showed that ICP-PHD could be able to track multiple targets effectively. Moreover, ICP-PHD spent less time compared with Gauss Hermite, and performed better in targets number estimation and state esti- mation comparing with eubature particle PHD （CP-PHD）.

关 键 词：多目标跟踪 概率假设密度 嵌入式求容积准则 Halton点集 

分 类 号：TN953[电子电信—信号与信息处理]