Fire Detection Berbasis Computer Vision Menggunakan YOLOv8 Secara Real-Time

Abstract

This study presents the development of a fire detection system using image processing techniques based on the YOLOv8 object detection algorithm to achieve fast, accurate, and real-time performance. A dataset of fire images with various visual characteristics was preprocessed, converted into YOLO annotation format, and used to train the model for 30 epochs. Evaluation results demonstrate that the YOLOv8 model performs effectively, achieving an mAP50 of 0.646, a precision of 0.889, and an inference speed of 282.5 ms per frame. The system is integrated with OpenCV to process webcam input and display bounding boxes and confidence scores in real time. The implementation confirms that YOLOv8 is a reliable solution for early fire detection, offering faster and more adaptive responses compared to conventional sensor-based methods. This approach can be applied to modern safety monitoring systems to enhance fire prevention efforts.