Fire Detection System with YOLO and Flicker Analysis Using HSV Color Space Masking

Early and reliable fire detection is crucial for minimizing fire damage and supporting firefighting efforts. Traditional sensor-based fire detection systems have limitations, prompting exploration into video analytics using computer vision techniques. This paper addresses the increasing fire incidents in Jakarta, Indonesia, underscoring the need for improved detection methods. We propose a video analytics-based fire detection system utilizing object detection with YOLO and flickering analysis with HSV color space. This approach aims to reduce false alarms by combining fire characteristics like flickering and color analysis, ultimately enhancing detection accuracy and contributing to improved fire safety measures. The proposed system achieved a 71.5% accuracy rate with 28.42 % fault detection rate, demonstrating proficiency in recognizing and differentiating real fire with static-fire objects, in order to reduce false alarms. Future development will focus on refining the fire detection algorithm to better capture the characteristics of moving fires and improving fire classification to reduce false positives, addressing the challenge of distinguishing real fires from highly realistic fire videos.