基于目标跟踪的粒子滤波重采样算法研究  

Resample Methodology in Particle Filters for Target Tracking

作　　者：袁韵洁[1] 张怡[1] 张玲玲[1] 

机构地区：[1]西北工业大学电子信息学院,陕西西安710072

出　　处：《计算机仿真》2010年第1期326-329,354,共5页Computer Simulation

摘　　要：传统粒子滤波(PF)中,重采样步骤里存在着粒子的"平均化"现象,导致粒子本身概率大小的因素被忽略,没有充分利用粒子集所包含的信息。通过改进抛弃小权值粒子的原则,以及充分利用粒子权值大小所代表的意义来进行粒子复制的两点进行算法改进,采用一维非线性目标跟踪模型和新的二维动态跟踪模型分别研究改进PF算法对于平均RMSE的影响。通过仿真,证明了改进后的算法可以显著降低变量的平均RMSE,特别是在二位动态跟踪模型中,使位置坐标和速度两种变量的平均均方根误差(RMSE)都有所改善,从而提高了滤波性能。In standard Particle Filtering（PF） algorithm,the ＂average problem＂ of particles replication in the resampling step neglects the significance of the probability of particles and does not sufficiently use the information contained in the particle pool.This paper improves the algorithm via reasonably setting the principle of abandoning small weight particles and replicating particles by the knowledge of weights.Experiments of one dimensional target tracing model and x-y target tracing model validated that the improved PF can ameliorate the average RMSE of variables.Especially in x-y target tracing model,better outcome can be got both in position and velocity.It is concluded that the improved algorithm can markedly reduce the RMSE of variables,and therefore the filtering performance of improved PF is greatly superior to that of the standard PF.

关 键 词：目标跟踪 粒子滤波 重采样 权值 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]