Distributed multisensor adaptive PMB filter with inaccurate measurement noise covariances

Abstract

In most multi-target tracking (MTT) scenarios, the prior information about the statistical characteristics of the measurement noise usually needs to be more accurate and pre-trained. However, in practice, this is often difficult to obtain accurately, especially when additive noise and multiplicative noise exist at the same time. In this case, the adaptive estimation of the measurement noise covariances is quite important for MTT. In this paper, an adaptive Poisson multi-Bernoulli (APMB) filter based on the random finite sets framework and variational Bayesian inference is proposed, which considers additive noise and multiplicative noise as a whole with a unified covariance. Moreover, the Gaussian inverse Wishart (GIW) implementation of the proposed APMB filter is given, where the Gaussian distribution describes the kinematics of the target, the IW distribution describes the covariance of the measurement noise, and the mixing of the GIW components in the Bernoulli components is calculated on average according to the weighted Kullback–Leibler. Furthermore, the proposed APMB filter is applied to a distributed multisensor, which can deal with inaccurate and inconsistent measurement noise covariances between different sensors. Finally, simulation results show that the proposed APMB filter can estimate the target effectively when the measurement noise covariances are inaccurate.