Detection and Geovisualization of Abnormal Vessel Behavior from Video

Abstract

Intelligent maritime situational awareness pursues an effective understanding of the majority of the activities related to the maritime domain (impacting the safety, security, economy, or environment), with the aid of artificial intelligence systems. Such an understanding requires the development of automated processes capable of not only detecting abnormal behavior but also of visually-representing and interpreting it. Although much progress has been made in anomaly detection and visualization using vessel self-reporting positioning data, there have been no corresponding advances using video data, despite the increasing use of cameras for maritime surveillance. In this work, we introduce a framework which goes beyond vessel tracking for anomaly detection in video, and is therefore applicable to scenes with a high density of vessels. The proposed framework detects abnormal behavior using a Generative Adversarial Network (GAN) and interprets this knowledge using metrics derived from clustering the positions and courses provided by an independent vessel/motion detector. These detections are geovisualized using an advanced displaying tool where detected abnormal behavior may be localized on the globe, providing an infrastructure for intelligent maritime situational awareness.