基于MCMC的改进粒子滤波算法  

Improved particle filter algorithm based on MCMC

作　　者：席志红[1] 付存利 

机构地区：[1]哈尔滨工程大学信息与通信工程学院,黑龙江哈尔滨150001

出　　处：《计算机工程与设计》2014年第9期3070-3074,共5页Computer Engineering and Design

基　　金：国家自然科学基金项目(60875025/f030410);中央高校基本科研业务费专项资金基金项目(HEUCF100818)

摘　　要：针对传统粒子滤波算法建议分布函数的选取问题和粒子退化现象,提出一种基于马尔可夫蒙特卡洛思想的改进粒子滤波算法。使用基于比例对称采样方法选取Sigma点的无迹卡尔曼滤波,产生粒子滤波并建议分布函数;将似然分布自适应权值调整策略应用于权值选取步骤;采用系统重采样方法,加入了用来保持粒子多样性的马尔科夫链蒙特卡洛步骤。仿真结果表明,该算法的估计状态能够更好地吻合真实轨迹,在非线性、非高斯场合的估计性能较优。To solve the problem of the proposal distribution function on selecting in the traditional particle filter algorithm and that of the particle degradation, an improvement particle filter algorithm based on Markov Monte Carlo ideas was proposed. Firstly, the unscented Kalman filter was applied that made use of the proportional symmetry sampling methods to emerge the Sigma points and to generate the recommendations distribution of particle filter. Secondly, the likelihood distribution adaptive strategy was.applied to the weights selection procedure; Finally, the system resampling methods and joining Markov chain Monte Carlo procedure were applied to maintain the particle diversity. Simulation results show that this algorithm can better match the real trajectory and has the merits of tracking accuracy in the non-linear and non-Gaussian occasion.

关 键 词：粒子滤波 目标跟踪 无迹卡尔曼滤波 蒙特卡洛 重采样 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]