Abstract: Detecting the anomaly events in maritime traffic is key to vessel maritime situational awareness. Automatic Identification System (AIS) data, initially envisioned for collision avoidance can also be used for detecting the anomaly vessel patterns due to its rich information content. The officers sitting at the Vessel Traffic Service (VTS), will monitor the vessel traffic and behavior of the vessels based on the received AIS data from the vessels. The VTS monitoring system receives large volumes of AIS data at the base stations, and manual detection of the anomaly patterns is nearly infeasible. To address the issue, we propose a novel approach based on the Local Outlier Factor (LOF) algorithm trained with customized features extracted from the AIS data to automatically detect the anomaly events during the vessel voyage. The proposed algorithm helps the officers to give attention only to the anomaly AIS instances, which will be comparatively less compared to the normal AIS instances. The difference in the statistical nature of the proposed extracted features from AIS data corresponding to normal and anomaly samples signified the relevance of the extracted features in anomaly detection. Experiments on a real-time dataset consisting of 1042002 AIS messages demonstrate the validity of the proposed method.

Key words: maritime surveillance, anomaly prediction, unsupervised learning, route estimation, contrario detection