Constrained adaptive Markov transition matrix based target tracking with IMMPF

Abstract

Interacting multiple models particle filter (IMMPF) has been recently paid great attention for its eminent ability in solving nonlinear target tracking problem. Through improving some filter steps, such as sampling and re-sampling, particle filter can offer more estimation accuracy. This paper proposes a particle filter taking advantage of constrained adaptive Markov transition matrix based on post-probability. At the end of each filter iteration process, we study two methods to update Markov transition matrix for the next iteration process. One is with the ratio of likelihood function, and the other is with the compress ratio of estimation error. Furthermore, to avoid possible failure resulted from abnormal data during the iteration process; we set the upper bound to constrain Markov transition probability. Simulations show that constrained adaptive Markov transition matrix is beneficial to improve interacting multiple models particle filter results.