基于均值迁移的粒子滤波算法研究

Particle filter algorithm based on mean-shift

针对弱观测噪声环境下的粒子退化现象,特别是观测噪声较小时非线性非高斯的粒子滤波问题,提出了一种基于均值迁移的粒子滤波算法。首先,将核密度估计的无参快速模式匹配算法引入到粒子滤波中,并迭代计算概率密度估计。然后,利用均值迁移估计粒子梯度的方向,计算每个粒子移向其样本的均值。当粒子位置发生改变时,对重采样粒子进行加权处理。最后,根据本算法采样更新粒子集,有效地克服了粒子退化现象并提高了状态估计精度。

To cope with particle degeneracy with weak measurement noise, especially the particle filter problems of nonlinear/ nonGaussian when the measurement noise is smaller, a particle filter algorithm is proposed based on meanshift. Firstly, nonparametric fast pattern matching algorithm of Kernel density estimation is introduced for particle filter, and the probability density estimation is iteratively calculated. Then, particle gradient direction and the mean value for each particle that moves to the sample are estimated by mean shift. When the position of particles is changed, the resampled particles are weightily pro cessed. Finally, using the method to update particle sets overcomes the particle degradation effectively and improves the accura cy of state estimation.