Robust visual-based tracking using deep learning with image enhancement for reach stackers in container terminals

Owing to limited visibility, container handling with stacking by reach stackers can lead to accidents and equipment damage under challenging weather scenarios. This study proposes a robust single object tracking (SOT) framework that integrates the actor-critic real-time tracking (ACT) algorithm with a dark channel prior (DCP) image dehazing module. This novel integration of a deep learning-based strategy effectively removes haze caused by adverse weather conditions, enhancing the performance of the object tracking model. Specifically deployed at Busan New Port (South Korea), this system provides reach stacker drivers with precise, real-time visibility via camera-based assistance in dynamic port environments. Evaluated on the container handling video custom dataset, the proposed method achieves a precision improvement of over 11 % compared to the ACT baseline algorithm, with the center location error (CLE) threshold at 20 pixels. By providing the driver with enhanced visual information, this solution minimizes downtime, enhances port productivity, supports smooth integration with port monitoring systems, and offers a practical tool for safer and more efficient container handling operations. • AI-powered software allows port authority to integrate computer vision into port business. • A novel scheme is developed for efficient object tracking in automated container terminals. • A proposed method employs deep learning-based detector with image enhancement. • The integrated scheme improves equipment handling with ensuring safety operations under challenging conditions. • With computer vision and AI technology, actor-critic tracking strategy creates a shipping industry leading cargo handling and service.