A Computationally Inexpensive Method for Anomaly Detection in Maritime Trajectories from AIS Dataset

Vessel behavior analysis can unfold valuable information about maritime situation awareness. Maritime anomaly detection deals with finding the suspicious activities of vessels in open water using AIS dataset. In this paper, an inexpensive method is introduced for automatic anomaly detection by utilizing historical sequence of trajectories of three vessel types of tanker, cargo and tug. We propose to project sequential data to a visual space in order to analyze and uncover the incongruent and inconsistent regions through segmentation of saliency maps. In order to evaluate the results, we take a statistical measurement for assigning the degree of anomality to each data point. The comparison of results indicate that this method achieves effective performance and efficient computational complexity in finding maritime anomalies in an unsupervised manner. In addition, this method is explainable and the results are visually interpretable.