Automated location of active fire perimeters in aerial infrared imaging using unsupervised edge detectors

A variety of remote sensing techniques have been applied to forest fires. However, there is at present no system capable of monitoring an active fire precisely in a totally automated manner. Spaceborne sensors show too coarse spatio-temporal resolutions and all previous studies that extracted fire properties from infrared aerial imagery incorporated manual tasks within the image processing workflow. As a contribution to this topic, this paper presents an algorithm to automatically locate the fuel burning interface of an active wildfire in georeferenced aerial thermal infrared (TIR) imagery. An unsupervised edge detector, built upon the Canny method, was accompanied by the necessary modules for the extraction of line coordinates and the location of the total burned perimeter. The system was validated in different scenarios ranging from laboratory tests to large-scale experimental burns performed under extreme weather conditions. Output accuracy was computed through three common similarity indices and proved acceptable. Computing times were below 1 s per image on average. The produced information was used to measure the temporal evolution of the fire perimeter and automatically generate rate of spread (ROS) fields. Information products were easily exported to standard Geographic Information Systems (GIS), such as GoogleEarth and QGIS. Therefore, this work contributes towards the development of an affordable and totally automated system for operational wildfire surveillance.