Video-Based Identification and Prediction Techniques for Stable Vessel Trajectories in Bridge Areas

Area of Science:

• Maritime safety
• Artificial Intelligence
• Vessel Traffic Management

Background:

• Vessel-bridge collisions are increasing globally, highlighting the need for better vessel track identification.
• Current methods using YOLO with DeepSORT/Bytetrack struggle with detection accuracy and boundary acquisition, leading to false positives and missed detections.

Purpose of the Study:

• To develop a novel framework for accurate vessel trajectory identification and prediction.
• To mitigate issues of false positives and negatives in vessel tracking.
• To enhance maritime safety through real-time anomaly detection and warnings.

Main Methods:

• Introduced Co-tracker, a long-term sequence multi-feature-point tracking method for accurate trajectory tracking.
• Utilized Long Short-Term Memory (LSTM) and Graph Attention Neural Network (GAT) for trajectory prediction and anomaly detection.
• Statistically calculated feature point cluster transformations for improved tracking accuracy.

Main Results:

• The proposed method demonstrates superior accuracy and captures essential heading angle features compared to YOLO with DeepSORT.
• Effectively reduced false positives and false negatives in vessel detection and tracking.
• Successfully provided real-time vessel group trajectories and predictions in the Ningbo Three Rivers area.

Conclusions:

• The novel framework significantly improves vessel traffic management efficiency and reduces collision risks.
• The system enhances secure navigation in complex, multi-target, and wide-area scenarios.
• Timely audiovisual warnings are issued for potential restricted zone entries, improving real-time alert functionality.