An Interactive Multiple Model Based Deep Learning State Fusion Approach to Target Tracking

The rapid development of the deep learning technology has provided novel, data-driven solutions to the classic maneuvering target tracking problems. Based on the celebrated interactive multiple model (IMM) approach, this paper proposes a deep learning state fusion target tracking algorithm, termed IMM-LSTM, which combines the advantages of the IMM algorithm and long short-term memory (LSTM) network. The algorithm assigns a separate target motion model for each of the LSTM trackers that are run in parallel. Further on, an LSTM-based classifier is employed to determine the weights for each motion model of the target and the final estimate is given by the weighted average of the estimates of these individual trackers. Simulation results have shown that our algorithm yields better tracking accuracy and robustness in scenarios where the a-priori target information is deficient.