Switch-Constrained Multiple-Model Algorithm for Maneuvering Target Tracking

Abstract

In most of the existing state estimation algorithms for maneuvering target tracking, the formulation of the mode switching process fails to accurately capture the deterministic law that the target mode will persist for a period of time once it turns to be in effect. In this paper, two strategies are presented to describe this characteristic. First, a constraint that model switching occurs no more than once over three consecutive time steps is proposed to provide a deterministic description that mode does not switch continuously. Second, considering target mode always lasts for some time while mode switching is completed instantly, the transition probabilities from one model to itself are fixed at extremely large values to enhance the certainty in mode sojourn segments. Based on these strategies, two recursive multiple model (MM) filters are derived in the framework of generalized pseudo-Bayesian (GPB) estimation and interacting multiple model (IMM) estimation, respectively. In both algorithms, although the large probability of model staying unchanged causes high peak errors at mode switching instants, the introduction of the switching constraint prevents the input to the mode-matched filter from being affected by the model transition probability, thereby obtaining a near-optimal performance in mode sojourn segments. Additionally, in order to reduce the peak estimation errors, a mode switching detector based on likelihood function is presented to work in parallel with the recursive MM filters. Once the detector declares that a mode switch occurs, the final estimate is adjusted from the output of the recursive MM filters to the estimate of the corresponding elemental filter, performance of which is guaranteed by the switching constraint. Simulation results demonstrate the capability of the proposed recursive MM filters to obtain performance close to that of a filter without model uncertainty in mode sojourn segments and the effectiveness of the mode switching detector to reduce peak errors at the same time.