YOLOv5-Based Detection of Electric Vehicles and Batteries in Residential Elevators for Fire Hazard Prevention

Electric bicycles (e-bikes) and their lithium-ion batteries have been identified as serious fire hazards when transported in confined spaces such as residential elevators. This paper proposes a YOLOv5-based object detection system to automatically identify electric vehicles (specifically e-bikes, also called “electrocars”) and standalone batteries in elevator camera feeds, enabling preventative safety measures. We present a dataset of elevator images collected from Chinese residential buildings with annotations for battery, electrocar, bicycle, and background. The YOLOv5 model is trained on this dataset and achieves high accuracy in detecting these objects. Experiments demonstrate robust performance: the model attains a mean average precision of 96.2 % across classes, with class-wise F1-scores around 0.9 at an optimal confidence threshold. We integrate various evaluation metrics including confusion matrix analysis and precision-recall curves to validate the system. Results indicate that YOLOv5 can reliably distinguish e-bikes and batteries even in the constrained elevator environment, with only minimal missed detections. This research highlights the potential of deploying real-time computer vision in elevators to automatically enforce fire safety regulations. The paper concludes with discussions on deploying such a system in practice and its role in fire hazard prevention in residential settings.