Fire and Smoke Detection Using the YOLOv8 Object Detection Model
DOI:
Keywords:
Fire detection, smoke detection, YOLOv8, object detection, GnConv, attention mechanism, ODConvBS, real-time surveillance.
Abstract
Fire and smoke incidents require rapid detection to reduce risks to people, property, and the environment. This paper presents a vision-based fire and smoke detection approach built around YOLO object detection, with an enhanced YOLOv8 configuration incorporating GnConv-based feature aggregation, attention mechanisms, and Omni-Dimensional Dynamic Convolution with Batch Statistics (ODConvBS). The system processes surveillance video through frame extraction and image pre-processing, then produces fire/smoke class labels, bounding boxes, and confidence scores. A Python implementation using PyTorch and the Ultralytics framework is exposed through a Flask-based inference application. The source study reports that a YOLOv5x6 baseline achieved 79.2% mAP, 80.6% precision, 74.0% recall, and 38 frames per second (FPS), while the enhanced YOLOv8 configuration achieved 89.4% mAP, 87.8% precision, 85.2% recall, and 46 FPS. These results indicate the potential of the enhanced detector for real-time visual monitoring. Dataset composition, training configuration, and test hardware must be documented in a reproducible evaluation
Downloads
Published
Issue
Section
License

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.


