RGB 채널 영상을 이용한 YOLOv8n 기반 실시간 화재·연기 감지 성능 평가 = Real-time Fire and Smoke Detection Performance Evaluation Based on YOLOv8n Using RGB Images

다국어 초록 (Multilingual Abstract)

Traditional sensor-based fire detection systems suffer from inherent structural limitations, including a high rate of false alarms caused by environmental factors and delayed initial responses due to the necessity of physical contact or close proximity to the ignition source. To address these critical issues, this study proposes and evaluates a real-time fire and smoke detection framework utilizing the YOLOv8n architecture based on RGB channel imagery. Comprehensive comparative experiments were conducted using the D-Fire dataset against the YOLOv5n and YOLOv8s models to verify performance metrics. The experimental results demonstrated that the proposed YOLOv8n model achieved the most optimal balance between detection accuracy and computational efficiency, recording an mAP@0.5 of 78.05% and a rapid inference speed of 51.1 FPS. Furthermore, with an extremely lightweight model size of approximately 6MB, the proposed system proved its viability for deployment on resource-constrained low-power edge devices without sacrificing performance. Consequently, this deep learning-based approach is expected to significantly enhance early fire response capabilities and situational awareness when integrated into future autonomous robotic surveillance applications.