Fire Detection and Notification System in Real-Time Using a Lightweight YOLOv8-Based Detector

In this paper, the FireDetection system is presented, which is designed to provide a robust and easily deployable solution for early fire detection. The system proposed uses the YOLOv8n detection model trained on an extended dataset, which includes detailed annotations for the presence of fire in each frame of the videos. The primary objective of this system is to furnish a resilient and readily deployable solution for early fire identification while simultaneously minimizing false positives and ensuring high frame rates to enable real-time applications. Through extensive experimentation, the results reveal significant enhancements in the model's performance, as evidenced by its analysis dataset accomplishments. With its robust architecture, advanced training, and reliable results, the FireDetection system emerges as a promising and valuable tool in the realm of fire prevention and early intervention, with potential applications in various scenarios that require swift and accurate fire detection. Finally, experimental results of this system demonstrate an improvement in state-of-the-art models.