An improved particle filter with a new sample processing strategy

In many nonlinear and non-Gaussian application areas, particle filter (PF) shows great promise. However, PF has a problem of sample impoverishment which decreases filter accuracy seriously. To improve the filter´s performance, this paper proposes an improved algorithm (PF-DR) with a new sample processing strategy. The new strategy introduces a desired region to select and adjust the particle samples. It accepts all the predicted samples which are already in the region, and concentrates the other samples which are not in the interval from remote region to high likelihood region of probability density function. Compared with standard PF, experimental results show that this new algorithm increases the filtering accuracy and reduces the total time consumption.