Flame filtering and perimeter localization of wildfires using aerial thermal imagery

Abstract

Airborne thermal infrared (TIR) imaging systems are being increasingly used for wildfire tactical monitoring since they show important advantages over spaceborne platforms and visible sensors while becoming much more affordable and much lighter than multispectral cameras. However, the analysis of aerial TIR images entails a number of difficulties which have thus far prevented monitoring tasks from being totally automated. One of these issues that needs to be addressed is the appearance of flame projections during the geo-correction of off-nadir images. Filtering these flames is essential in order to accurately estimate the geographical location of the fuel burning interface. Therefore, we present a methodology which allows the automatic localisation of the active fire contour free of flame projections. The actively burning area is detected in TIR georeferenced images through a combination of intensity thresholding techniques, morphological processing and active contours. Subsequently, flame projections are filtered out by the temporal frequency analysis of the appropriate contour descriptors. The proposed algorithm was tested on footages acquired during three large-scale field experimental burns. Results suggest this methodology may be suitable to automatise the acquisition of quantitative data about the fire evolution. As future work, a revision of the low-pass filter implemented for the temporal analysis (currently a median filter) was recommended. The availability of up-to-date information about the fire state would improve situational awareness during an emergency response and may be used to calibrate data-driven simulators capable of emitting short-term accurate forecasts of the subsequent fire evolution.

A Bi-Spectral Microbolometer Sensor for Wildfire Measurement

“…In an effort to expand the scientific and operational value of infrared (IR) imaging, there is a growing body of evidence suggesting that it can be used to characterize wildfire behaviour under experimental conditions. Specifically, through the automation of fire structure detection [ 14 , 15 ] and temporal analysis, it is possible to map and measure the rate and direction of spread [ 16 , 17 , 18 ], arguably with higher precision than traditional methods [ 19 ]. Additionally, the fire radiative power (FRP; W) [ 20 ] emitted from actively spreading flame fronts has been shown to be a strong predictor of fire line intensity (kW m −1 ) [ 21 , 22 ].…”

Inverse modelling in wildfire spread forecasting: towards a data-driven system

“…This process was manually performed as expert judgement was the best available tool at the moment. Recently, advances in automatising this filtering methodology that showed high agreement with the manually performed step (Valero et al, 2015(Valero et al, , 2016.…”

On the use of compact thermal cameras for quantitative wildfire monitoring

“…Some variables that can be estimated remotely using IR cameras are the fire perimeter location, the rate of spread, the Fire Radiative Power (FRP) and Byram's fire intensity (Pérez et al 2011;Paugam et al 2013;Stow et al 2014;Butler et al 2016;Dickinson et al 2016;Johnston et al 2017). In this regard, our previous work contributed to automate the analysis of aerial TIR imagery through computer vision techniques (Valero et al 2017(Valero et al , 2018. Algorithms of this type allow the automated acquisition of spatially explicit, high-frequency fire information, and they may be the base of a quantitative monitoring system with applicability to wildfire incident management and wildfire research.…”

“…In these cases, unrealistic ROS values may be produced and they need to be detected and rejected. There are image processing techniques to achieve this by directly filtering flames out (Valero et al 2017), but such methods are complex to implement and need considerable computational resources. Sometimes it is enough with the application of common statistical rules for outlier rejection.…”