Computer vision for wildfire research: An evolving image dataset for processing and analysis

Tom Toulouse (1) , Lucile Rossi (1) , Antoine A Campana (1) , Turgay A Celik (2) , Moulay A Akhloufi (3)

1 Feux
2 WITS - University of the Witwatersrand [Johannesburg]
3 University of Moncton

The last decade has witnessed the use of computer vision for wild-fire detection and measurement. The first and most important step for computer vision analysis is the fire pixel detection because it determines the accuracy of the following processing. The evaluation and the comparison of the wildfire detection algorithms of the literature and the development of new ones needs open datasets with a large number of annotated images and their ground truth. We address this issue by presenting a publicly evolving wildfire annotated image database with ground truth data with examples of use. Currently, it contains 500 visible images and, in a more limited number, multimodal images and videos with frame by frame annotations. This is currently the largest dataset released in this research field.

Mots clés: Database; image; wildfire; pixel detection

Domaines: Traitement des images [eess.IV]; Vision par ordinateur et reconnaissance de formes [cs.CV]; Base de données [cs.DB]

Tom Toulouse, Lucile Rossi, Antoine A Campana, Turgay A Celik, Moulay A Akhloufi. Computer vision for wildfire research: An evolving image dataset for processing and analysis. Fire Safety Journal, 2017, 92, pp.188-194. ⟨10.1016/j.firesaf.2017.06.012⟩. ⟨hal-01560570⟩