Our logo!
Our overall flow!

Project Story: Wildfire Detection with Live Camera Feeds

About the Project

Our project aims to revolutionize wildfire detection by integrating live California forest cameras with fine-tuned machine learning models. By automating the process of identifying wildfire outbreaks in real-time through a custom ML model trained and deployed on SageMaker and MLFlow, we provide critical tools for early detection, leveraging a robust AWS and Databricks ecosystem to warrant swift action and protecting lives, ecosystems, and infrastructure.

Inspiration

With the recent California wildfires devasting the land, we decided that there had to be something that could help improve the situation. early detection has become a top priority. Monitoring live feeds from forest cameras is labor-intensive and prone to delays. We were inspired to develop a solution that combines the power of computer vision and machine learning to detect wildfires faster and more reliably than manual methods.

What We Learned

Throughout the development journey, we gained valuable insights into various AWS services and machine learning techniques. Not only that, but we were introduced to the power that Databricks has to offer with our utilization of MLFlow and Databricks Notebooks for our model. Alongside this, we further deepened our understanding of AWS Lambda and EventBridge for scheduling tasks, AWS SageMaker & API Gateway for model training and deployment, and Python + React TypeScript for the backend and frontend work

Building the Project

Our project centered around React TypeScript on the frontend and Python + AWS + Databricks to process and manage the data passed into our frontend. The overall AWS/Databricks ecosystem centered around AWS Lambda and EventBridge which triggered a scheduled function to scrape the latest contracts from the DOD website. These contracts were then fed into our custom NLP model (trained via MLFlow & Databricks Notebooks) and deployed on AWS SageMaker via API Gateway .

Once analyzed, the statistics regarding the fire probability of the current camera is fed into our front end for display and analyzation. If there is a high fire probability chance, we also proceed to store the current snapshot of the camera and alert users who are tracking the cam of a potential fire risk on our dashboard.

Challenges Faced

One of the main challenges we faced while creating our project was figuring out how to effectively utilize Databricks in our overall flow. We were initially very unfamiliar with Databricks tools and were unsure of how it could be integrated with AWS or our current project.

Accomplishments that we're proud of

Trained, fine-tuned, and deployed a transformer based CV model to effectively analyze images for wildfire risk.
Developed a working Databricks/AWS ecosystem to automatically parse and analyze live camera footage of over 100+ cameras capturing the California wilderness. -Created a sleek UI to view insights into our CV model.