Maneuvering Target Tracking Based on Hybrid Inference

This paper proposes a hybrid inference method combined with sequential Monte Carlo and variational Bayes, to approximate the intractable posterior distribution. The joint problem of nonlinear state estimation and model parameter identification is transformed into the optimization problem of maximizing the lower bound of evidence. Under the framework of sequential Monte Carlo, a parameterized variational proposed distribution is first designed, and then a closed-loop iterative mechanism of sampling particle weight and proposed distribution hyperparameters is constructed. The joint optimization of target state estimation and model parameter identification is carried out by stochastic gradient descent. Finally, the estimation accuracy of our proposed method is super to the state-of-art methods by simulated data verification.