Adaptive distributed multiple-model filter with uncertainty of process model

Abstract

Target tracking is a critical application in multiple-sensor state estimation. However, it is subject to uncertainty in the process model due to target maneuvering, which includes motion model uncertainty and process noise uncertainty. In this paper, we present an adaptive distributed multiple-model filter for maneuvering target tracking that solves the two types of uncertainties by variational Bayesian. Initially, a centralized algorithm is developed over multiple-sensor. Using Categorical distribution and inverse Wishart distribution as the conjugate prior of motion model and process noise covariance respectively, the target state, model identity and noise parameters are jointly estimated by variational Bayesian. The algorithm is then extended to a distributed version where prior information and likelihood function achieve consensus through the Alternating Direction Method of Multipliers (ADMM). We also analyze the consensus and convergence of the proposed algorithm. The simulation results demonstrate the effectiveness of the proposed algorithm for maneuvering target tracking under process model uncertainty.