Fire and smoke detection with high accuracy using YOLOv5

Abstract

Early detection of fire and smoke is crucial for preventing catastrophic losses in various environments. This research presents a novel approach to fire and smoke detection using motion estimation algorithms integrated with the YOLOv5 object detection framework. The proposed method leverages the temporal characteristics of fire and smoke propagation to enhance detection accuracy and reduce false positives. We introduce a multi-stage pipeline that combines optical flow-based motion estimation withYOLOv5’s real-time object detection capabilities.The system is evaluated on a diverse dataset of fire and smoke scenarios, demonstrating significant improvements in detection speed and accuracy compared to traditional methods. Our results show a 15% increase in mean Average Precision (mAP) and a 30% reduction in false positive rates, making this approach promising for real-world applications in fire safety and surveillance systems.

Keywords: Machine learning, CNN architectures, Motion estimation, Fire detection, Smoke detection, Machine vision