Anomaly detection in maritime data based on geometrical analysis of trajectories

Anomaly detection is an important use of the Automatic Identification Systems (AIS), because it offers support to users to evaluate if a vessel is in trouble or causing trouble. For instance, it can be used to detect if a ship is doing something that may cause an accident or if it has changed its route to avoid bad weather condition. In this work, a new method for finding anomalies in the ships' movements is proposed. The method analyzes the trajectory of ships from a geometrical perspective. The trajectory of the ship is compared with a near-optimal path that is generated by a graph search algorithm. The proposed method extracts some scale-invariant features from the real trajectory and also from the optimal movement pattern, and it compares the two sets of features to generate an abnormality score. The method is unsupervised and it does not require training. Instead of labeling the trajectories as normal/abnormal it calculates a score value that denotes the extent of abnormality. The scoring scheme provides a ranking system in which the user can sort the trajectories based on their abnormality score. This is useful when dealing with large number of trajectories and the user wants to picks the most abnormal cases. For the evaluation, the method was run on three months data of North Pacific Ocean and score values were generated. Among the entire dataset, 100 randomly chosen trajectories were labeled by an expert. After applying a threshold on the score value, the proposed method had 94% accuracy.