Every Vessel Counts: Neural Network Based Maritime Traffic Counting System.

Area of Science:

• Computer Vision
• Maritime Technology
• Artificial Intelligence

Background:

• Conventional maritime traffic monitoring systems like AIS and VTS lack comprehensive data, particularly in diverse port environments.
• Mediterranean ports face challenges with existing systems due to varied vessel types and operational complexities.
• Accurate maritime traffic data is crucial for efficient port management and in-depth maritime research.

Purpose of the Study:

• To develop and evaluate a real-time vessel counting system using land-based cameras for enhanced maritime traffic monitoring in ports.
• To improve data acquisition for maritime research by overcoming limitations of existing systems.
• To provide a robust and accurate solution for vessel detection and classification in complex maritime settings.

Main Methods:

• Implementation of a YOLOv4 Convolutional Neural Network (NN) trained on the novel SPSCD dataset for vessel classification into 12 categories.
• Utilizing a Kalman tracker combined with the Hungarian Assignment (HA) algorithm for multi-target vessel tracking.
• Incorporating a stability assessment to mitigate false positives from non-vessel objects and improve tracking reliability.

Main Results:

• The system demonstrates an average counting accuracy of 97.76%.
• Achieved an average processing speed of 31.78 frames per second, indicating high efficiency.
• Captured 386% more maritime traffic data compared to conventional Automatic Identification System (AIS) methods.

Conclusions:

• The proposed camera-based system offers a fast, robust, and effective solution for real-time maritime traffic monitoring.
• The system significantly enhances data collection capabilities, providing valuable insights for maritime research and port operations.
• This technology presents immense potential for improving the comprehensiveness of maritime traffic data analysis.