Maritime anomaly detection based on statistical methodologies: theory and applications

The proposed research aims at contributing to advances in the anomaly detection methodologies within the framework of maritime domain, in order to improve the ability to reveal, understand, anticipate and prevent illegitimate activities at sea. This work has been developed based on three fundamental tools: a prior information from a maritime traffic graph that can be derived from a route atlas or from historical data, the Ornstein-Uhlenbeck mean reverting stochastic process to model the vessel's dynamics in deep waters, and the complete or incomplete observation of the available data from heterogeneous sensor systems. Relying on the statistical hypothesis testing framework, the work treats the problem of detecting a vessel's anomalous deviations from the expected conditions in the presence of different levels of data unavailability. The problem is further complicated by the possible falsification of dynamic data self-reported by the vessel. A worst-case scenario in terms of detection capability is finally tackled by proposing an optimization methodology to make the trajectory of a malicious vessel as stealth as possible. The effectiveness of the proposed strategies has been assessed through experimental analyses concerning both synthetic and real-world maritime operational scenarios.