An improved tracking Kalman filter using a multilayered neural network

This paper presents a method for improving the estimation accuracy of a tracking Kalman filter (TKF) by using a multilayered neural network (MNN). Estimation accuracy of the TKF is degraded due to the uncertainties which cannot be expressed by the linear state-space model given a priori. The MNN capable of learning an arbitrary nonlinear mapping is thus added to the TKF to compensate the uncertainties. The MNN is trained so that it realizes a mapping from, the measurements to the corrections of estimations of the TKF. Simulation results show that the estimation accuracy is much improved by using the MNN.