Smart Gas Cylinder Warehouse Management Using Computer Vision for Automated Inventory, Classification, and Safety Monitoring

Abstract

This research applies the YOLOv8 model, running on local edge computing, for real-time gas cylinder detection and classification. The system was trained to identify and distinguish gas types (classification) and to analyze the physical state of the cylinders (posture analysis) in two forms that can lead to warehouse hazards or accidents: tilted and fallen. The technical performance evaluation revealed that the developed model achieved an mAP@ 0.5 of 93.8% for overall gas cylinder detection. Moreover, its safety monitoring performance demonstrated a recall rate of 99.0% for detecting fallen/tilted cylinders. This system is integrated with the WMS (warehouse management system) via an API service running in a cloud container to automatically update cylinder locations and stock counts. It also provides alerts if cylinders are placed in the wrong zone. The results demonstrate the potential to reduce stock-taking time by 4 5. 5 5 %, increase inventory accuracy by 4 9. 0 1 %, and sustainably elevate safety standards within the warehouse. These contributions support the advancement of industrial innovation and smart infrastructure (SDG 9) while enhancing workplace safety and operational efficiency (SDG 8).