By Paolo Franchi
Adaptive multisensor-multitarget tracking with belief propagation. Soldi, Giovanni; Meyer, Florian; Braca, Paolo. CMRE-FR-2017-008. December 2018.
Multisensor-multitarget tracking is a challenging task which aims at estimating time-varying states of moving objects by exploiting the measurements from multiple sensors. These returns are usually corrupted by noise, with missed detections and false alarms usually modelled in a statistical fashion, where often model parameters, i.e. detection probabilities, clutter rates or motion model parameters, are supposed to be known a priori and not changing in time. However, in real applications such parameters are typically unknown and thus have to be estimated adaptively in order to avoid performance degradation. This report considers an adaptive Bayesian multitarget-multisensor tracking framework in which the tracking problem is formulated according to the measurement origin uncertainty paradigm and is able to react and adapt to the time-varying changes of various unknown parameters. The resulting Bayesian estimation problem is efficiently solved using the belief propagation (BP) message passing scheme. A concrete example is reported and described, where the dynamics of the targets follow multiple dynamic models and the target detection probabilities at each sensor are unknown. The same approach can be applied for the adaptive estimation of other parameters that are usually unknown and possibly time-varying, e.g. the clutter intensity. The performance of the proposed adaptive BP method is validated by means of several numerical simulations and by testing it in a real case scenario with measurements from two High-Frequency Surface Wave radar systems.