Abstract:

A new filtering method, called as the Rao-Blackwellized convolution filtering (RBCF) algorithm, is presented for conditional linear Gaussian state space models. The method uses a convolution filter (CF) to estimate the nonlinear states of the model while its linear states are estimated by the Kalman filter (KF). The new algorithm, which uses CF based on the kernel function, improves the precision of estimation compared with the RaoBlackwellized particle filter (RBPF) when the noise is low. The simulation results of the proposed method applying to tracking the maneuver target show that the estimate precision of the RBCF is obviously higher than the RBPF,  and the root mean square error (RMSE) of position and velocity is lower by more than an order of magnitude comparing with RBPF, when the noise is low. The tendency is more obvious as the noise become lower.