ABSTRACT Accurate lane line segmentation is critical for auto‐gantry steering of Rubber‐Tyred Gantry (RTG) cranes in automated container terminals. RTG lanes are typically straight but often exhibit degradation due to surface wear, contamination or interference from nearby ground markings, making pixel‐level detection particularly challenging. This letter proposes a lightweight and robust segmentation algorithm based on an enhanced BiSeNet V2. Improvements include depthwise separable convolutions to reduce model complexity, efficient channel attention to enhance feature discriminability under contamination and weak contrast, and an edge‐aware loss function to address the thin‐line nature of RTG lanes. Experiments on a curated set of labelled RTG images demonstrate that the proposed method outperforms existing lightweight segmentation models in accuracy and robustness. The model achieves a favourable balance between precision and computational efficiency, making it suitable for real‐time use in port automation systems. To facilitate future research and reproducibility, the annotated RTG lane dataset used in this work is publicly available at: https://github.com/DONN‐RD/Lane‐line‐detection .
Dong et al. (Thu,) studied this question.