一种适用于低测量噪声系统的粒子滤波算法

An Particle Filter Algorithm for the Low Measuring Noise System

针对常规粒子滤波算法使用先验密度函数来采样粒子,从而使粒子分布依赖动态模型来降低估计精度的问题,以基于观测量相似函数采样的相似采样粒子滤波为基础,提出一种改进的粒子相关性预采样相似采样粒子滤波算法.在系统测量噪声较小的情况下,利用相似采样获得更加贴近真实后验分布的粒子来提高估计精度;而相关性预采样则通过计算相邻时刻粒子的转移概率并淘汰概率较低的粒子来提高粒子利用效率,在保证估计精度的同时显著降低粒子数量需求.设计了算法的重要性密度函数并推导了权值递推公式.通过蒙特卡洛仿真分析了本文提出的算法.最后通过一个混合坐标系下的目标跟踪实例阐述了算法的应用.

Aiming at the problem that conventional particles filter algorithm uses a prior density function to sample particles, thereby the particles distribution should rely on the dynamic model to reduce the estimation precision. A improved particles correlated pre-sampling likelihood sampling particles filter algorithm is proposed, which based on the likelihood sampling particles filter of observation likelihood function sampling. Under the condition of low measurement noise, the likelihood sampling can obtain particles which are closer to the true posterior distribution, so the estimation precision is expected to be improved. The correlated pre-sampling procedure calculates the transition-probability of adjacent time and abandons the particles with lower probability to improve particles efficiency. By this way, estimation accuracy is ensured and the amount of required particles is decreased significantly. The importance density function is designed and the weight-value recursive formula is deduced. The proposed algorithm is analysised by the Monte Carlo simulation, and it is also applied to the problem of target-tracking in the hybrid coordination.