Real-time Container Tracking and Damage Detection at Seaports Using Deep Learning

Abstract

Efficient container handling and early damage detection are critical for minimizing operational delays, reducing costs, and ensuring safety in global maritime logistics. This work presents a deep learning-based methodology for real-time container tracking and automated damage detection during crane unloading operations at container terminals. We develop and deploy two specialized YOLOv12-based object detection models: one for identifying containers in motion and another for detecting structural damages such as bents, dents, and holes. Our models are trained and evaluated on a real-world dataset curated from video feeds captured at the EUROGATE Container Terminal in Limassol, Cyprus. The system is designed for robust performance under realistic terminal conditions, including variable lighting and motion. Our models achieve high detection accuracy, with a mAP50 of 0.99 for container detection and 0.75 for damage detection, substantially outperforming existing benchmarks. These results highlight the practical potential of our method for improving efficiency and safety in automated maritime logistics.