Development of Real Time System for Smoke and Fire Detection in Wide Areas Using Yolov8

Abstract

In an era marked by significant advancements in Artificial Intelligence (AI) and its diverse applications across various fields, including machine learning and computer vision, the domain of surveillance systems and safety measures has undergone a profound transformation. Amidst numerous natural disasters, fires emerge as one of the most devastating calamities, necessitating the utilization of AI capabilities to develop intelligent monitoring systems that bolster our defensive efforts against this disaster. This involves the early detection and notification of relevant authorities to prevent irreparable damages. Traditional fire detection devices, while yielding satisfactory results, exhibit diminished effectiveness in open or large areas and lack real-time detection capabilities. In response to these challenges, this study aims to develop an advanced real-time smoke and fire detection system specifically designed for wide deployment. Leveraging the capabilities of the YOLOv8 deep learning model, the study trained the most suitable versions of the proposed model (YOLOv8l, YOLOv8m) with varying hyperparameters on a dataset comprising 9756 images of various smoke and fire scenarios. The results demonstrate the models' capability to accurately detect fires and smoke, achieving commendable average precision rates while maintaining a delicate balance between precision and recall. Specifically, the YOLOv8l model achieved a mean average precision (mAP50) of 85.1% and an F1 score of 80%, while the YOLOv8m model achieved a mAP50 of 86% and an F1 score of 82%. These models exhibit promising results in real-time fire and smoke detection systems, indicating a new era of proactive measures for fire detection and prevention, deployable on unspecified specification surveillance cameras.