Multisensor Target Tracking and Fusion With ECM: A Unified Message-Passing Method

Tracking multiple maneuvering targets in complex environments with false alarms and electronic countermeasures (ECM) poses a significant challenge due to the intractable high-dimensional inference. This article presents a unified message-passing (MP) approach that converts the inference problem into an optimization framework based on probabilistic graphical models, providing an approximate solution for joint state estimation and environment parameter identification. The unified MP approach integrates the mean-field approximation and belief propagation to enable efficient, scalable, and flexible inference. Conjugate priors and the mean-field update rule are used to analytically approximate the posterior probability density functions (PDFs) for continuous latent variables such as target kinematic states and ECM parameters. For discrete latent variables, such as target visibility, motion mode, and data association, the belief propagation update rule is used for its compatibility with hard constraints. Finally, the posterior PDFs of high-dimensional latent variables are iteratively updated to effectively address the coupling issue between estimation and identification. By directly processing raw measurements from multiple sensors, our proposed method exhibits a measurement-level fusion architecture, which significantly enhances target tracking performance in complex ECM environments. Simulation results show that the proposed method effectively tracks multiple maneuvering targets with range gate pull-off interference.