Detecting illegal diving and other suspicious activities in the North Sea: Tale of a successful trial

摘要

While the volume of data available to surveillance systems is steadily increasing, the assessment of a complex situation is still mainly done by humans. This might result in wrong assessments or even in entirely missing crucial situations. An automated system to assess incoming data and highlight situations of interest could greatly support human experts to perform their tasks. This work describes an approach to model expert and domain knowledge in order to reason about situations of interest. The model is a Multi-Source Dynamic Bayesian Network (DBN), which encodes different situations and their relationships at different levels of abstraction. Through the use of Bayesian inference, the existence probability for a situation can be estimated. This paper reports on a successful implementation of a multi-source DBN to detect illegal diving activities, tested during live sea trial in the North Sea under the EU project MARISA. During the trial the Netherlands Coastguard performed different manoeuvres simulating anomalous behaviours, including illegal diving. The multi-source DBN Behavior Analysis service processed real data streams of AIS and radar contacts and positively detected the unique event, early, without false alarm.