Fire and Smoke Detection
The Fire and Smoke Detection tool aims to recognise automatically the presence of fire and smoke textures in visual content. Visual input can be provided in the form of images or videos for on-demand analysis, but the tool can also process video streams to support online monitoring of an area. The tool is also equipped with an alert notification mechanism, and can be integrated within a crisis management solution broadcasting alerts in case fire or smoke are detected. A visualization of the analysis outcome is provided bellow. The red and grey boxes indicate the presence of fire and smoke in the respective video snippets, as detected by the tool.
The tool uses image classification techniques on still images or sequential video frames to estimate the presence of the interested events. The analysis is based on state-of-the-art image classification, inspired from the success of deep learning in image understanding and scene recognition. Specifically, the VGG-16  model was initially pre-trained on the Places365 dataset , in order to initialize its features for generic scenery understanding. Then, the model was fine-tuned in a compilation of fire and smoke public datasets [3, 4, 5, 6, 7]. The compiled database contains more than 20000 images overall, of which 4500 are related to fire and 1430 are related to smoke. The tool can reach processing speed up to 50fps depending on the resolution of the input (image or frame) and the deployment infrastructure.
Minimum System Requirements
Supported OS: Linux/Windows
Processor: Quad Core CPU
Hard drive: 50GB of free space
Graphics: NVIDIA GPU with 8GB VRAM
 Simonyan, Karen, and Andrew Zisserman. “Very deep convolutional networks for large-scale image recognition.” arXiv preprint arXiv:1409.1556 (2014).
 Zhou, Bolei, et al. “Learning deep features for scene recognition using places database.” (2014).
 Chino, Daniel YT, et al. “Bowfire: detection of fire in still images by integrating pixel color and texture analysis.” 2015 28th SIBGRAPI conference on graphics, patterns and images. IEEE, 2015.
 Cazzolato, Mirela T., et al. “Fismo: A compilation of datasets from emergency situations for fire and smoke analysis.” Brazilian Symposium on Databases-SBBD. SBC, 2017.