[Paper Review] Composite Event Recognition for Maritime Monitoring

Manolis Pitsikalis, Alexander Artikis+4 more|arXiv (Cornell University)|Mar 7, 2019

Data Management and Algorithms21 references4 citations

TL;DR

This paper presents a real-time composite event recognition system for maritime monitoring using the Run-Time Event Calculus (RTEC) to detect complex vessel activities. It introduces a formally specified library of 25 maritime patterns—validated with domain experts—enabling accurate detection of behaviors like trawling, piracy, and illegal fishing, with sub-second processing on large-scale AIS data streams.

ABSTRACT

Maritime monitoring systems support safe shipping as they allow for the real-time detection of dangerous, suspicious and illegal vessel activities. We present such a system using the Run-Time Event Calculus, a composite event recognition system with formal, declarative semantics. For effective recognition, we developed a library of maritime patterns in close collaboration with domain experts. We present a thorough evaluation of the system and the patterns both in terms of predictive accuracy and computational efficiency, using real-world datasets of vessel position streams and contextual geographical information.

Motivation & Objective

• Develop a formal, declarative system for detecting complex maritime events using real-time AIS data streams.
• Address the challenge of identifying dangerous, suspicious, and illegal vessel activities in dynamic maritime environments.
• Collaborate with domain experts to define a comprehensive library of actionable maritime patterns beyond basic anomalies.
• Evaluate system performance on real-world datasets for both predictive accuracy and computational efficiency.
• Provide publicly available event recognition results to support future research in pattern learning and maritime situational awareness.

Proposed method

• Employ the Run-Time Event Calculus (RTEC) as the underlying formalism for composite event recognition with declarative, logic-based semantics.
• Integrate spatial preprocessing to compute real-time spatial relations (e.g., proximity, within-area) between vessels and geographical regions.
• Apply trajectory simplification to filter non-critical AIS messages, reducing data volume while preserving event-relevant signals.
• Construct a hierarchical library of 25 maritime patterns (e.g., trawling, tugging, SAR) using RTEC’s fluent and event primitives, with dependencies on lower-level patterns.
• Use sliding window processing over time to detect temporal event sequences, with deadline mechanisms to manage event termination.
• Evaluate performance on two real-world AIS datasets: 18M messages (Brest, 2015–2016) and 55M messages (Europe, 2016), enriched with static geographical data (e.g., Natura 2000 zones, fishing areas).

Experimental results

Research questions

• RQ1 How can composite event recognition be formally specified to detect a wide range of complex maritime activities in real time?
• RQ2 What is the impact of data compression (via critical point filtering) on the accuracy of event detection in AIS streams?
• RQ3 How efficient is the system in processing large-scale, real-time AIS data streams while maintaining high detection accuracy?
• RQ4 Can expert-validated maritime patterns be effectively encoded and recognized using a logic-based event calculus framework?
• RQ5 To what extent does the use of contextual geographical information improve the detection of illegal or suspicious vessel behaviors?

Key findings

• The system achieves real-time composite event recognition, processing a 16-hour window of over 50,000 critical AIS events in less than one second on a single-core desktop.
• Critical point stream compression reduced input volume significantly, but introduced false negatives and positives due to message removal, particularly affecting trawling and SAR pattern detection.
• The most computationally intensive patterns were trawling and search-and-rescue (SAR), due to RTEC’s deadline mechanism for event termination.
• The library of 25 maritime patterns, developed in collaboration with domain experts, enables detection of diverse activities including illegal fishing, piracy, and unusual routing.
• All recognized composite events from the Brest dataset (18M messages) were made publicly available to support future research in pattern learning and anomaly detection.
• The system demonstrated robust performance across both terrestrial and satellite AIS data, especially in delayed data environments common in maritime monitoring.