Enhacing Logistics with Computer Vision and Fog Computing-Driven Auto-ID Technologies

In this paper we propose a series of enhancements to a computer vision system specifically designed for auto-identification (Auto-ID) of loads in industrial and logistic settings using a fog computing architecture. The system effectively monitors multiple docks within a logistics warehousing management system, identifying loads on pallets in real-time and providing immediate feedback to ensure proper truck loading. Its Auto-ID functionality has been expanded to include visual estimation of the class and approximate size of pallets, resulting in an efficient method for automatic image labeling that enables the creation of few-shot training datasets. Our experimental results show that these datasets are sufficient for leveraging on state-of-the-art foundational computer vision models to obtain high accuracy and precision rates on both object classification and load size estimation by precise bounding box placement. Furthermore, our evaluation also demonstrates that employing simple techniques like linear probing significantly reduces the need for extensive processing and cloud-based training, perfectly aligning with the system’s underlying fog computing architecture. This streamlined approach simplifies parameter calibration, eliminates the need for extensive data uploads, and facilitates flexible online updates as new load types and class samples are integrated into the system.