Bayesian estimation via extended and unscented Kalman particle filtering for non linear stochastic systems
Faycal Souibgui, Fayçal Ben Hmida, Abdelkader Châari · 2013
State estimation is of paramount importance in many fields of the problems encountered in practice. Filtering is the method of estimating the sate of the system by incorporating noisy observations. Particle filters are sequential Monte Carlo methods that use a point mass representation probability densities in order to propagate the required statistical proprieties for state estimation. In this paper, a new formulation of particle filter for nonlinear Bayesian estimation frameworks using various proposal importance function densities and state characterizations. New formulation particle filtering methods that use the extended and unscented Kalman filters are introduced. All the methods are compared in terms of accuracy and robustness. Is proposed from the sequential Bayesian approach theory. A synthetic stochastic model that incorporate non-linear, non stationarily is used for illustrative example.