Analysis of a new nonlinear filter and tracking methodology (Corresp.)

A recursive nonlinear filter and tracking methodology is developed for a class of partially observable processes with an approximating model which is linear in the unobservable states and initially has the unobservables conditionally Gaussian with respect to the observations. The usual model smoothness is not required, and applications to simulated tracking problems show the filter to be considerably more accurate than the modified second-order filter which in a general sense includes the extended Kalman filter.