Real-time long short-term glance-based fire detection using a CNN-LSTM neural network.

Vision-based fire detection is widely studied recently to reduce the damage of fire disaster thanks to the advantages of software-based methods comparing to traditional hardware-based fire detection using sensors. This paper presents a novel method for fire detection using the convolutional neural networks on image sequences of videos to extract both the spatial and temporal information for fire classification. The system includes a CNN network to extract the image features, and short-term and long-term stages at the end for classification. Experiments carried out on the common public datasets show promising results in terms of performance in comparison to the previous works.