基于随机摄动再采样的粒子概率假设密度滤波器  

Particle Probability Hypothesis Density Filter Based on Stochastic Perturbation Re-sampling

作　　者：徐从安[1] 何友[1] 夏沭涛[1] 程俊图 董云龙[1] 

机构地区：[1]海军航空工程学院信息融合研究所,烟台264001 [2]中国人民解放军91213部队,烟台264000

出　　处：《电子与信息学报》2016年第11期2819-2825,共7页Journal of Electronics & Information Technology

基　　金：国家自然科学基金(61471383;61304103)

摘　　要：作为概率假设密度滤波的典型实现方式,粒子概率假设密度滤波器无需线性高斯等先验假设,因而在多目标跟踪中得到了广泛的应用。为解决粒子退化问题并保持粒子规模,该滤波器引入了重采样机制,然而,该重采样机制易引起粒子多样性耗尽,导致粒子贫化问题产生。为解决这一问题,该文提出一种新的基于随机摄动再采样的粒子概率假设密度滤波器。首先,全面分析了粒子概率假设密度滤波因粒子贫化问题导致目标失跟的过程。然后设计了一种随机摄动再采样算法,该算法在重采样导致粒子多样性缺失时,根据源粒子的位置与复制次数随机产生相应数目的新粒子,并对源粒子进行删减,其可在保留源粒子信息的前提下保持粒子的多样性。最后,该文将该算法纳入概率假设密度滤波框架,提出了一种新的粒子概率假设密度滤波器。仿真结果表明该滤波器在不显著增加运行时间的前提下能够克服粒子贫化问题,相比标准的粒子概率假设密度滤波器具有更好的跟踪性能。As a typical implementation of the Probability Hypothesis Density （PHD） filter, Particle PHD （P-PHD） is suitable for highly nonlinear systems and widely used in Multi-Target Tracking （MTT）. However, the resampling in P-PHD filter, recommended to avoid particle degeneracy, introduces the problem of diversity loss among the particles, namely particle impoverishment problem. To solve the problem and improve the performance of the P-PHD filter, a novel filter based on stochastic perturbation re-sampling is proposed. First, a comprehensive analysis on the particle impoverishment problem of P-PHD filter is presented. Then for the purpose of keeping the particle diversity, a new stochastic perturbation re-sampling algorithm is developed, which generates new particles according to the position and duplicating times of the original particles, and removes some excessive copied particles. Finally, the re-sampling algorithm is integrated into the P-PHD filter framework and a Stochastic Perturbation Particle PHD （SPP-PHD） filter is proposed. Numerical examples show that the proposed filter can overcome the particle impoverishment problem and improve the estimation performance on the premise of not significantly improving the simulation time.

关 键 词：多目标跟踪 概率假设密度 粒子滤波 随机摄动再采样 

分 类 号：TN391[电子电信—物理电子学]