Stochastic filtering is presented as the elementary theory of particle filter, and its defects which lead to particle regression when applied in particle filter, are analyzed. Following this characteristic, some improved techniques are introduced. Compared their advantage and disadvantage when used them in particle filter, then give their common problem that is sampling difficulty when used in high-dimensional state space sampling. Against this problem, a new structured particle filter named as iterated particle filter is proposed, and do simulation to initial alignment for SINS using EKF and EK-PF, so as to explain origin of this new particle filter.
© 2012 Optical Society of AmericaPDF Article