Fixed-interval retrodiction approach to Bayesian IMM-MHT for maneuvering multiple targets

In a Bayesian framework, we propose a hierarchy of suboptimal retrodiction algorithms that generalize Rauch-Tung-Striebel (RTS) fixed-interval smoothing to multiple hypothesis tracking (MHT) applications employing interacting multiple model (IMM) methods (IMM-MHT). As a limiting case we obtain new simple formulae for suboptimal fixed-interval smoothing applied to Markovian switching systems. Retrodiction techniques provide uniquely interpretable and accurate trajectories from ambiguous MHT output if a certain (small) time delay is tolerated. By a simulated example with two maneuvering targets that operate closely spaced under relatively hard conditions we demonstrate the potential gain by fixed-interval retrodiction and provide a quantitative idea of the achievable track accuracy and mean time delay involved