摘要
扩展卡尔曼粒子滤波(EPF)在预测阶段通过EKF选取重要性函数而优化了粒子选取,但是传统EPF算法中粒子权值一般是通过正态分布的概率密度函数计算的。此方法没有突出不同噪声粒子的权值差别,在计算中引入了较大的相对误差。通过在更新阶段对权值计算所依赖的概率密度函数做出改进,得到改进的EPF算法。同时采用实际目标跟踪数据进行仿真对比实验,结果验证了此方法有效可行,并且减小了预测误差。
Extended Particle Filter（EPF） determines the importance function based on EKF in the predicting stage and optimizes the choice of particles.In traditional EPF algorithm,however,the particle weight is usually obtained using normal distribution function as probability density function.The weight differences among particles with different noises are not emphasized,which may bring large relative error in the calculation.We made improvement to the probability density function upon which the weight calculation is dependent in the stage of update,and established an improved EPF.Simulation was made in tracking system using real data.It turned out that the improved EPF is available and effective,which can also decrease the prediction error.
出处
《电光与控制》
北大核心
2011年第4期10-12,25,共4页
Electronics Optics & Control
基金
东北电力大学博士基金(BSJXM-200802)
吉林省教育厅项目(吉教科合字2009100)
吉林市科技发展计划资助项目(20090601)
关键词
目标跟踪
粒子滤波(PF)
EPF
概率密度函数
权值计算
正态分布
反比例函数
target tracking
Particle Filter（PF）
EPF
probability density function
weight calculation
normal distribution
inverse proportional function
