[Tesis postgrado] [Pagada] Deteccion de Fuego en la naturaleza usando IA

Early wildfire detection is of the utmost importance to enable rapid response efforts, and thus minimize the negative impacts of wildfire spreads. To this extent, we propose to install a networks of stations composed of cameras connected to Raspberry Pi that process the images in real time in order to automatically detect smoke plumes using Computer Vision algorithms. We scrapped the web in order to create a new database of smoke plumes’ sequence of images (videos).

The challenges are various:

• The detection is currently tackled using a classical state-of-the-art object detection model (Yolov8) that do not take into account the sequentiality
• The images are processed on a light computer, this makes space to work more frugal models
• A benchmark of the SOTA models is needed
• How to improve the quality of the dataset by using bigger models offline (even though they cannot be used online)
• Improve the model for early detection (an exemple)

This is a project in collaboration with the non-profit association PyroNear and the Corporacion Nacional Forestal.