Real-Time Fire Detection Using YOLOv8 and Twilio SMS Alerts

Abstract

Fire alarm systems are essential components of modern safety infrastructure, playing a critical role in mitigating risks to human life, property, and the environment. Despite their necessity, traditional sensor-based methods—such as smoke and heat detectors—often struggle with significant limitations, including delayed response times and high rates of false alarms. These issues are particularly prevalent in large industrial warehouses or open-air environments where environmental variables can impede the accuracy of physical sensors.

To address these challenges, this study presents a real-time fire detection system that integrates the YOLOv8 deep learning architecture with the Twilio SMS communication service. The system utilizes the YOLOv8 (You Only Look Once) algorithm, which is highly regarded for its ability to perform high-speed and accurate object detection in video streams. The model was trained using a diverse and comprehensive dataset consisting of 11,263 annotated images from Roboflow, ensuring reliable detection across various lighting and environmental conditions.

The platform continuously analyzes live video feeds to identify the visual signatures of fire. Upon confirmed detection, it automatically triggers an instantaneous SMS alert via the Twilio API to pre-defined emergency recipients. Quantitative evaluation demonstrates the model's high efficacy, achieving a mean Average Precision (mAP@.5) of 95%, with a precision of 95.4% and a recall of 76%. With an average confidence level of 0.58%, this system offers a robust, automated solution that bridges the gap between detection and emergency notification.