Supervised learning for adaptive interactive multiple model (SLAIMM) tracking

Abstract

To improve target tracking algorithms, supervised learning of adaptive interacting multiple model (SLAIMM) is compared to other interacting multiple model (IMM) methods. Based on the classical IMM tracking, a trained adaptive acceleration model is added to the filter bank to track behavior between the fixed model dynamics. The results show that the SLAIMM algorithm 1) improves kinematic track accuracy for a target undergoing acceleration, 2) affords track maintenance through maneuvers, and 3) reduces computational costs by performing off-line learning of system parameters. The SLAIMM method is compared with the classical IMM, the Munir Adaptive IMM, and the Maybeck Moving-Bank multiple-model adaptive estimator (MBMMAE).