Fire is a serious hazard in different places around the world. The detection of fire attains greater importance in the last decades due to the loss and damages caused by fires. Thus, A fire detection system is required for minimizing injuries with financial loss. Previous approaches to fire detection using deep learning have relied on a Convolutional neural network (CNN), which have limitation in terms of speed and real-time detection. This paper proposes a custom YOLOv5 fire detection system providing high accuracy and improved real-time performance. This work represents a real-time video fed into a deep learning model and this approach shows promising results in detecting the fire with high accuracy of 98.3%, precision of 98.6%, F1– score of 96% with low false positive rate. The system is deployed on a Raspberry Pi 3 Model B for efficient and low-cost implementation, providing a timely warning to prevent property damage and loss of life.

Fire detection, YOLOv5, Real-time video, Raspberry pi, Deep learning