A new sampling method in particle filter

This paper presents a new method to draw particles for the particle filter in the case of large state noise. The standard bootstrap filter draw particles randomly from the prior density which does not use the latest information of the observation. Some improvements consist in using extended Kalman filter or unscented Kalman filter to produce the importance distribution in order to move the particles from the domain of low likelihood to the domain of high likelihood by using the latest information of the observation. The performances of these methods vary with the structure of the models. We propose a modified bootstrap filter which uses a new method to draw the particles. Our method outperforms the bootstrap filter with the same computational complexity. The effectiveness of the proposed filter is demonstrated through numerical examples.