Real-time detection of fires and smoke in healthcare facilities using advanced deep learning models on live video streams of surveillance cameras

Fires in healthcare facilities pose a critical risk to patients and staff, yet conventional detection systems often respond slowly and cannot operate in real time. This study proposes an AI-driven fire and smoke detection system leveraging existing CCTV networks and advanced deep learning models for rapid hazard recognition.We focus on YOLOv11, the latest in the YOLO object detection family, and benchmark its performance against YOLOv8. A custom dataset of 17,525 images with 27,314 annotated fire and smoke instances was compiled, encompassing diverse indoor and outdoor scenarios. All YOLOv11 variants (nano to extra-large) were trained and evaluated, achieving high detection accuracy, with the medium model reaching a mean average precision (mAP@50) of 90%. The results highlight how model size affects detection speed and accuracy, demonstrating the feasibility of deploying AI-based real-time fire detection systems in healthcare environments to enhance safety and minimize false alarms.