Tom Toulouse

-Detection of fire pixels in the visible and near infrared spectra.

-Multimodal stereovision for fire pixel detection.

-Development of a multimodal image database and tests for the various algorithms.

-Search for matching points on pairs of images of different modalities.

-Three-dimensional data cross-matching.

-Estimation of the geometric characteristics of a fire front using 3D points.

-Algorithm tests for fires spreading across 100 square meters.

-International collaboration with the MIVIM of the Université Laval, the CRVI of Lévis and the University of the Witwatersrand in Johannesburg.

The author of 2 international publications, she has taken part in 1 international conferences.

• T. Toulouse, L. Rossi, T. Celik, M. Akhloufi, X. Maldague, "Benchmarking of wildland fire color segmentation algorithms", IET Image Processing, 2015

• T. Toulouse, L. Rossi, T. Celik, M. Akhloufi, "Automatic fire pixel detection using image processing: A comparative analysis of rule-based and Machine learning-based methods" Signal, Image and Video Processing, 2015

• L. Rossi, T. Toulouse, M. Akhloufi, A. Pieri, and Y. Tison, "Estimation of spreading fire geometric characteristics using near infrared stereovision", Proc. IS&T/SPIE Electronic Imaging SPIE 2013, "3D Image Processing (3DIP) and Applications 2013", p. 86500A 86500A, March 2013

• T. Toulouse, L. Rossi, M. Akhloufi, A. Pieri, and Y. Tison, « Stéréovision proche infrarouge pour l’estimation de caractéristiques géométriques de feux de végétation », Techniques, mesures et systèmes, Université Laval, 2013

• L. Rossi, D. Cancellieri, J. L. Rossi, F. Morandini, M. Akhloufi, and T. Toulouse, « Intégration de la vision dans une recherche pluridisciplinaire dédiée aux feux de forêt », CNR IUT, Corte, France, 2013

I joined the UMR CNRS 6134 Science for the Environment fire team as part of the internship for the second year of my Masters "Signal and Image Processing" at the University of Bordeaux. My internship focused on a calculation of the geometric characteristics of a spreading fire based on a system of near infrared stereovision. I opted for an internship in Corsica for a number of reasons. The first was the appealing internship topic both in terms of research into vision and in terms of its application. The second reason was the magnificent work setting in Corte.

Fire spread experience during my internship

During my internship, supervised by Lucile Rossi, we demonstrated that near infrared allowed for a much better view of the fire, particularly in the presence of fire. At the end of my internship, I had the opportunity to further my research through a thesis under joint supervision with the Université Laval in Québec under the guidance of Lucile Rossi and Xavier Maldague. The Université Laval is unique in that it has a research chair specialised in infrared vision. During my stay in Canada, I was able to study the various spectral ranges of infrared for fire vision. During my thesis, we developed a mechanism drawing on two systems of multimodal stereovision (capturing images of visible spectrums and infrared simultaneously). Initially I primarily worked on the detection of fire pixels in the various spectral ranges (visible and infrared) and in particular, I developed a method simultaneously using visible and infrared images to accurately detect the presence of fire in the image concerned. With the aim of evaluating the performance of detection algorithms, we compiled a database of fire images containing fire-related information (ground truth, main colour, presence of smoke, brightness of the environment, and other factors). This database is now available online. I also worked on cross-matching data from the two systems to a common reference point so as to have a scatter plot representing the fire. Geometric characteristics such as the height, length, speed, surface area, scope and volume of the fire are subsequently calculated automatically based on the scatter plot representing the fire. To test the system with its related programs, we conducted several outdoor experiments mainly on fires spreading across a 50m² surface area.

Experience at valley level

The results of this thesis demonstrated that the use of multimodal stereovision systems allows for both the effective detection and a good three-dimensional reconstruction of the fire.

Three-dimensional reconstruction of fire based on programs developed in the thesis as applied to the various system images

I defended my thesis in November 2015 and then furthered my research as a Temporary Lecturer and Research Assistant at the University of Corsica, working primarily on the transfer of my programs so that they could be used on a drone for Vito Cuillo’s thesis.