Assessment of vessel route information use in Bayesian non-linear filtering

Bayesian non-linear filtering is considered in this paper for the state vector estimation of manoeuvring targets at sea. Innovative schemes based on the Extended Kaiman Filter and the Particle Filter are derived by the introduction of a priori vessel route information. Such contextual input drives the selection of the manoeuvre model to be used for target state prediction. This aims at coping with significant measurement gaps suffered by coastal sensors - due to their limited spatial coverage or temporal revisit. The capabilities of the context-aided techniques are assessed for realistic scenarios that include typical vessel manoeuvres. The Kullback-Leibler Divergence is adopted as performance metric. The analysis demonstrates that the use of the a priori information yields dramatic improvements in highly non-linear conditions for target tracking, and the Particle Filter outperforms the Extended Kaiman filter approach in the exploitation of the route information.