摘要
粒子滤波算法摆脱了解决非高斯滤波问题时随机量必须满足高斯分布的制约,近年来广泛应用于跟踪与定位研究中。与粒子滤波有关的一个普遍问题是退化现象,增加粒子个数可以部分的解决这个问题,同时马尔可夫链的引入可以使粒子分布更加合理,因此建议分布的选择是至关重要的。分析粒子滤波原理后,将马尔可夫链蒙特卡罗法方法引入粒子滤波算法的实现中,结合扩展卡尔曼滤波和不敏卡尔曼滤波两种建议分布进行仿真。仿真结果展示了改进的粒子滤波算法的良好性能,而且粒子退化现象得到有效遏制。
Particle filter overcomes the restriction that stochastic variant must be Gaussian when dealing with the non - Gaussian filtering problem, and is widely used in the tracking and locating recently. The unavoidable degeneracy problem in particle filter can be solved to some extent by increasing the number of particle. At the same time, using Markov Chain can make the particle distribution more reasonable, so the selection of the proposal distribution is crucial. In this paper, after analyzing the principle of particle filter, MCMC is brought forward to realize this method and combined with two kinds of proposal importance distribution （ EKF and UKF） to carry out the simulation. The simulation results indicate the good performance of these improved particle filters and the degeneracy problem is avoided efficiently.
出处
《计算机仿真》
CSCD
北大核心
2009年第4期120-124,共5页
Computer Simulation
