A novel multiple-model treatment for maneuvering target tracking

This paper addresses the maneuvering target tracking (MTT) problems for linear discrete-time dynamic systems with multiple models. A generalized formulation of the multiple-model (MM) dynamic systems is proposed, and a new treatment called random parameters Kalman filtering (RPKF) is investigated. The random parameters including some coefficient matrices and noise covariances are used to characterize the motion mode uncertainty of the MTT problem at each time instant. The proposed approach is different from the classical MM approach which has been received great attentions for solving the MTT problem. Compared with the interacting multiple model (IMM) algorithm that uses model probabilities to weight the inputs and outputs of a bank of parallel Kalman filters, the RPKF algorithm uses model probabilities to weight the possible coefficient matrices and noise covariances resulting in performing a single filter. The recursive state estimation formulae for a class of specific MM dynamic system are provided for three scenarios which are optimal in the mean squared error sense under some conditions. Some numerical simulations demonstrate the performance, including the good estimation accuracy and low computational complexity, of the proposed RPKF algorithm by comparing with the IMM algorithm.