Container Detection from Gantry Cranes Exploiting an Auto-Prompting SAM*
Yunjian Feng, Jun Li · 2024
Container detection is essential for automated container loading and unloading. Small and medium-sized terminals increasingly opt for vision solutions over expensive LiDAR solutions to automate their rubber-tired gantry cranes for container handling. However, most existing image processing-based methods fail to meet robustness and real-time requirements in practical scenarios. On the other hand, although deep learning-based object detection approaches enhance adaptability to complex environments, their reliance on large-scale datasets hinder their applications. This paper presents an approach to container detection for rubber-tired gantry cranes, exploiting an Auto-Prompting Segment Anything Model (AP-SAM). Initially, an Auto-Prompter is proposed for the first time. It strengthens the similarity between reference and target features by training a feature adaptor module that uses only a single reference image with segmentation labels. Next, local peak points of the similarity map are selected as prompts, guiding the SAM to achieve container segmentation. Experimental results show that our proposed method can achieve a container detection precision of 89.62% and a F1-score of 87.16%.