Task 2: Methods for anomaly and collision detection
Under the scope of Task 2, a data mining approach has been developed for probabilistic characterization of maritime traffic and ship anomaly detection (Rong et al. 2020). The approach automatically groups historical traffic data provided by the Automatic Identification System (AIS) in terms of ship types, sizes, final destinations and other characteristics that influence the maritime traffic patterns off the continental coast of Portugal. Then, relevant waypoints along a route are identified where significant changes in the ships’ navigational behaviour are observed using trajectory compression and clustering algorithms. This provides a vector-based representation of the ship routes consisting of straight legs and connecting turning sections that facilitates route probabilistic characterization and anomaly detection. The maritime traffic from the traffic separation scheme off Cape Roca to the ports of Lisbon, Setúbal and Sines is characterized probabilistically at the identified route legs and waypoints in terms of lateral distribution of the trajectories and speed profile. Ship trajectory anomaly detection is then achieved based on the developed maritime traffic probabilistic models.
An approach to assess the ship-ship collision probability of encounter scenarios taking into account the uncertainty in the trajectories of the ships along specific routes has been developed (Rong et al. 2019). The approach is based on the probabilistic trajectory prediction model that describes the uncertainty in future positions of the ships by continuous probability distributions represented by a Gaussian Process. The collision probability is formulated as the percentage of the ship trajectories that lead to collisions, assessed by samples of smooth continuous trajectories generated from the Gaussian Process model. The approach was demonstrated in several encounter scenarios and ship routes derived from historical trajectory information provided by Automatic Identification System data.
