Wildfire-Watch

A scalable wildfire fire detection framework, written in Python.

Getting Started

Installation

Conda (Recommended)

conda env create -f requirements.yml
conda activate WFW

Pip

pip install -r requirements.txt

Flags

Wildfire-Watch has five flags a user can define, listed as follows:

video - A path to an OpenCV video input, which is set to 0 for a local webcam
capRate - Sets a capture rate. For every specified number of frames, a Google Vision API call will be made
info - determines if detection information will be logged to the console
output - determines if forest fire detections will be saved as individual images
Display - determines if each analyzed frame is displayed, for debugging purposes

Input

An input video must be supplied for Wildfire Watch to analyze. Every 100 frames of the video will be run through Google Cloud VISION to search for wildfires.

Output

You can find the output detection(s) of forest fires in the detections folder. Each detection is formatted as follows:

FRAMENUMBER_LATITUDE_LONGITUDE.png 

where FRAMENUMBER is the current frame number. LATITUDE and LONGITUDE denote the latitude and longitude generated from an IP address.

Resulting Images

You can find the outputted detection image(s) in the detections folder.

Wildfire Detected

No Wildfire Detected

Developer Notes

All object identification uses Google Cloud's VISION API. The pipeline can be exported across multiple systems with minimal changes.

Sample pipeline

To get started, specify a camera input using the flags in main.py, then run the excecutible using python ./main.py. You can view the frames determined to contain wildfires in them in the detections folder.

Wildfire-Watch is Open Source. You can find the code on our GitHub repository.