Particle Filter Based on Adaptive Part Resampling
-
摘要: 针对传统重采样算法易引起粒子贫化的问题,提出了自适应不完全重采样粒子滤波 (A particle filter based on adaptive part resampling, APRPF)算法. APRPF以分步的方式仅对部分粒子进行重采样,以递推的方式计算表征 粒子退化程度的度量函数(Measurement of particle degeneracy, MPD),直到满足给定条件.重采样后的粒子由新生粒子 和未参与重采样的粒子组成,前者的存在有助于缓解退化问题,后者可使粒子集保 持一定多样性.实验结果表明,与标准粒子滤波(Sampling importance resampling, SIR)、辅助变量粒子滤波(Auxiliary particle filter, APF)、正则化粒子滤波(Regularized particle filter, RPF) 三种滤波器相比, APRPF的估计精度高;由于平均重采样次数少,计算量也小.Abstract: A particle filter based on adaptive part resampling (APRPF) is proposed to solve the problem of particle impoverishment introduced by traditional resampling algorithm. In APRPF, only small portion of samples are resampled in a step by step manner, and a recursive formula is developed to evaluate the measurement of particle degeneracy (MPD). The resampling process continues until MPD satisfies the given condition. After resampling, the particle set consists of two subsets, one contains new born particles, and another contains unresampled particles. The former can help to alleviate particle degeneracy, whereas the latter is in favor of improving the diversity of particles. Experimental results show that APRPF has less computational cost and more precise filtering results than sampling importance resampling (SIR), auxiliary particle filter (APF) and regularized particle filter (RPF).
-
Key words:
- Particle filter (PF) /
- sample degeneracy /
- sample impoverishment /
- part resampling
点击查看大图
计量
- 文章访问数: 2155
- HTML全文浏览量: 94
- PDF下载量: 1308
- 被引次数: 0