Nonlinear interacting particle filter algorithm

In state estimation problem of nonlinear non-Gaussian systems, the analytical form of the posterior density function is hard to gain, so the common particle filter employs state transition density function as importance proposal distribution without considering the latest observation. For the above problem, a nonlinear interacting multiple model method is developed. The method is used to generate the importance density function (importance proposal distribution), based on which a modified particle filter, nonlinear interacting particle filter, is proposed. The new importance proposal distribution takes the latest observation into considerations, which makes it much more close to the posterior density function. Experiments show the effectiveness of the proposed algorithm.