YOLOv11-Based Real-Time Fire and Smoke Detection for Automated Surveillance

The early and accurate detection of fire and smoke is critical for minimizing risks in surveillance applications, yet traditional systems often face limitations in speed and visual identification. This research addresses these challenges by developing and evaluating a high performance, real time fire and smoke detection system for automated surveillance, focusing on the advanced YOLOv11 deep learning architecture. The model was trained and validated on a comprehensive dataset of 12,614 images, consisting of 8,080 fire and 4,534 smoke instances sourced from public repositories. Data augmentation and preprocessing techniques were employed prior to training for 50 epochs on an NVIDIA RTX 3090 GPU. The resulting system, implemented in Python with OpenCV, integrates an automatic alarm notification feature for practical deployment. Performance evaluation demonstrates the system’s high efficacy, achieving a mean Average Precision (mAP@0.5) of 96.3%, a precision of 96.5%, and an inference speed of 55 frames per second (FPS). While testing confirmed stable detection across diverse environmental conditions, challenges remain in identifying small scale objects such as thin smoke or partially obstructed flames. The findings conclude that the YOLOv11 based approach provides a robust and efficient solution for automated fire hazard detection, significantly outperforming baseline models and affirming its suitability for deployment in critical real world environments.