BUILDING A REAL-TIME FIRE DETECTION SYSTEM

Abstract

Fire hazards pose significant risks to human life, property, and infrastructure, particularly in residential areas, factories, and industrial facilities. Conventional fire alarm systems, which rely primarily on smoke or heat sensors, often exhibit delayed response times and lack the ability to provide visual information about the actual situation at the incident site. These limitations reduce the effectiveness of early warning and may lead to slower intervention during emergencies. Therefore, the development of computer vision–based fire detection systems has become increasingly essential to enhance real-time monitoring and safety management. In response to this demand, this thesis focuses on designing a Real-Time Fire Detection System utilizing deep learning techniques, with the YOLOv11 architecture chosen for its high detection accuracy and rapid processing speed. The system can analyze live video streams from surveillance cameras and identify fire occurrences as soon as the visual signal is received. This computer vision–driven approach effectively addresses the drawbacks of traditional sensor-based systems while providing direct visual evidence to support timely decision-making and emergency response. The system follows a client–server architecture, in which the backend— implemented using the Flask framework—handles video processing, performs inference using the YOLOv11 model, and manages communication between modules. All fire detection events, including timestamps, confidence scores, and captured frames, are stored in a PostgreSQL database for monitoring and later analysis. A user-friendly web interface built with HTML, CSS, and JavaScript allows operators to view real-time video streams, access detection history, and manage camera sources efficiently. Additionally, the system integrates an automatic email notification feature to ensure immediate alerts when potential fire incidents are detected. Experimental results indicate that the system operates reliably, achieves high detection accuracy, and meets real-time performance requirements. With these advantages, the proposed solution contributes a practical and effective approach for fire safety monitoring, offering strong potential for deployment in residential, industrial, and public environments.