自适应交互多模型的PHD粒子滤波多机动目标跟踪  

PHD Particle Filter Method for Tracking Maneuvering Targets Based on Adaptive Interactive Multiple Models

作　　者：危璋 冯新喜[1] 毛少锋 

机构地区：[1]空军工程大学信息与导航学院,西安710077

出　　处：《弹箭与制导学报》2015年第2期166-170,共5页Journal of Projectiles,Rockets,Missiles and Guidance

摘　　要：针对多机动目标跟踪中采用统一固定模型转移概率的问题,提出一种在线估计模型转移概率的自适应多模型PHD滤波(AIMM-PHD)。首先保留模型的采样粒子及其似然度;其次根据粒子的分类结果,计算出每个目标对应每个模型的状态输出;最后将输出交替作为模型输入进行滤波,计算出目标的模型转移概率。实验表明:相较于IMM-PHD,所提AIMM-PHD有较低的OSPA误差,目标个数估计更准确,且时间只增加了8.1%,从而证明了该算法的有效性。To solve the problem that multiple model probability hypothesis density （ IMM-PHD） filter for maneuvering target tracking uses the prior model transition probability, a adaptive algorithm to Markova transition probability proposed. Firstly, the particles and the likeli-hood every model in the process of particles interaction, and then the output of every model to every target according to assortment in the process of state estimation, lastly, the model transtions probability by Bayes principle. The results show：compared IMM-PHD, AIMM-PHD has lower OSPA error;higher accuracy of target number estimation but its time only increases 8. 1%, thus the effectiveness of the proposed algorithm.

关 键 词：概率假设密度 交互多模型 自适应模型转移概率 

分 类 号：TN953[电子电信—信号与信息处理]