Wildfire Watch

GIF
Wildfire Watch correctly detecting wildfires.
GIF
Wildfire Watch avoiding false-positives on a clean video.

Inspiration

Throughout 2020, there have been multiple devastating wildfires all around the world. Headlines filled our newsfeeds, ranging from infernos racing their way across Australia to gender reveal parties sparking flames in California. We knew that there needed to be a better way of managing how wildfires are detected. Our shared passion for environmentalism, coupled with our emerging skills in Python and interests in artificial intelligence, inspired us to take the first steps to a cleaner future. We decided that to solve the problem, it would be essential to detect wildfires as early as possible, and in turn, to respond to them quickly.

What it does

Wildlife watch starts with OpenCV reading frames from a deployed drone video feed. The video feed is continually passed into Google’s vision AI, which returns a list of detected phenomena with associated probabilities. If our pipeline detects a high chance of a wildfire, the frame is saved and the instance’s current geolocation is stored. This information can be used to alert relevant authorities.

How we built it

Wildfire Watch was created using Python, OpenCV, Google Cloud, and Google’s Vision API.

Challenges we ran into

It was all of our team members’ first interaction with Google Cloud computing services, and it was quite a bit of a learning curve to understand what to do and why we were doing it. A great deal of struggle was had in crossing that first boundary; however, the API proved exceptionally useful, and we managed to pull through.

Accomplishments

We came up with a novel, plausible solution to a problem that plagued our social media and morning news. To do so, we implemented Google Cloud Vision AI for real time wildfire detection, and implemented real-time geolocation tracking from an instance’s IP address.

What we learned

Most members of our team had our first experience in AI/ML working on this project. We all learned about the Google Cloud platform, and about utilizing Google’s Vision API. In addition, many of us had never worked on such a collaborative Python programming project as this, so we built up our collaboration and communication skills. We also learned how to implement multiple APIs in conjunction on a large scale project.

What's next for Wildfire Watch

Seamless integration with live video feed.
Cross correlation of fire location across multiple drones
Ability to directly notify and partner with local law enforcement and fire departments.
Pathfinding AI for drones, synchronous movement between drones to enable wide and continuous coverage.

Built With

Submitted to

Knight Hacks
- Winner Technology as a Force for Good from PWC

Created by

I researched the problem and was in the group working with OpenCV to capture frames. In the past I used OpenCV to implement template matching so I expanded my knowledge in the library. This was my first time working with Google Cloud, AI/ML and with python on a large scale. This is also my first Hackathon so it was very intimidating but also exciting and offered a lot of knowledge!

Natalie Ruiz-Sanchez
I helped in setting up the Google Cloud project, and conducted research for and brainstormed the functionalities of our idea. I headed the design and presentation of our product. In this first interaction with artificial intelligence and machine learning as well as Google Cloud, I picked up new skills in a language I am still new to, Python.

Brian Tieu
I lead the backend development of this project. I helped implement OpenCV, Google’s Vision API, and geolocation tracking into our framework. I learned how to use multiple API’s on a large-scale project, and I also learned how to cross-reference geolocation from an IP address.

Ammaar Firozi
I helped set up the Google Cloud project and the Google Vision connection. It was my first time working with Google Cloud, my first experience with AI/ML, and my first time working on a larger scale python project.

Nicholas Yardich