2.4 KiB
Fire Detection
Abstract
Forest fires represent an important threat to natural ecosystems. Early detection is essential to prevent extensive damage and reduce risks associated with them. Conventional fire detection systems primarily rely on smoke or temperature-based sensors. These approaches have inherent limitations that restrict their effectiveness, particularly in outdoor environments and forested areas.
Our approach attempts to combine the feed obtained from traditional video or surveillance cameras with a motion detection algorithm and a model to predict forest fires in real-time. This approach is more cost-effective and requires less human intervention than other methods.
Implementation
We use a Motion Detection Algorithm that uses a background subtraction method with recursive updates of thresholds and estimated backgrounds.
The used model is an XGBoost model pre-trained on labeled image datasets containing fire and non-fire images. A histogram was computed for the color channels to differentiate without using the entire images.
Additional details can be found in the Technical Report and in the Project Slides.
Installation
First, create a new conda environment using:
conda env create -f environment.yml
Activate the new environment using:
conda activate FireDetection
Usage
The file main.py in the src directory contains a simple interface for using the program.
To execute, use:
python ./src/main.py
To change the video that is being used for detection, replace the video_path value in line 18 of main.py with the path to your target video.
Additionally, the algorithm can be tested with a live feed from http://66.119.104.155/mjpg/video.mjpg.
The example video was sourced from: Fire: Fountaingrove in Santa Rosa (Monday, Oct. 9)
Authors
This project was made as part of the 2023 AI Lab: Computer Vision and NLP course at Sapienza Università di Roma.