A network abstraction of multi-vessel trajectory data for detecting anomalies

Abstract

The detection of anomalies in vessel trajectories is a problem of great interest for all maritime surveillance systems, since it may uncover strange, suspicious or difficult situations for vessels. All the existing works in the field examine specific aspects of the problem and propose case specific tools that can hardly generalize or scale-up to a worldwide monitoring system. In this article, we present a methodology for creating a network abstraction of the trajectories of multiple vessels, which uses only the information collected from the vessels’ Automatic Identification System (AIS). The resulting network abstraction contains rich information about the vessel behavior in an area and can be processed with network analysis and other data mining techniques in order to uncover hidden outliers, even in an unsupervised manner. Experimental results on a real dataset demonstrate some of the capabilities of the proposed network model and indicate its extension to more complex automatic surveillance tasks.