Detecting Anomalous Vessel Trajectories: A Collaborative Clustering-Based Approach

Thousands of vessels operating worldwide may present issues related to anomalous trajectories, often characterized by unexpected changes in course, speed, or navigation through restricted areas. Manually detecting these anomalies is impractical, evidencing the need for automated support for maritime surveillance agents. Although Automatic Identification Systems (AIS) enhance situational awareness, they are insufficient for detecting trajectory anomalies. Existing approaches already use AIS data to detect anomalous vessel trajectories, but they do not consider contextual variables (e.g., oil spills) or the domain expertise of maritime surveillance agents. This paper introduces INSTRUCTOR, a solution that combines AIS data with clustering algorithms to group vessel trajectories based on similar directions and movement patterns. This clustering output assists experts in collaboratively identifying anomalous trajectories, a process validated through the collective judgment of multiple domain specialists. Experiments conducted in partnership with the Brazilian Navy corroborate the effectiveness of INSTRUCTOR in detecting anomalous maritime trajectories.