En-vision

Front Page
Zoomed in front page
Katmai National Park main page
Computer Vision and general data
Daily auto updating animal detection graphs
Computer Vision and general graphs
Yellowstone demo main page
example of ML computer vision model
example of ML computer vision model
Yellowstone demo data

Inspiration

Teammates first hackathon so we let him choose --- I want to get into animal and ecological data science or software engineering. My grandpa was a wildlife photographer, and because of him I've always been interested in animals. I've decided to combine my interest in animals with my interest in computer science and machine learning. My end goal is to work for a company that helps with wildlife conversation or tracking through software engineering. I would love to work on things like bioacoustics or computer vision for animals.

What it does

By using publicly available cameras from parks, we can monitor what and when animals are seen to provide tourists with a predictable account of the animals they are likely to see while visiting. Additionally, through the social features of the app, tourists can notify other users of animals they saw that are hidden from cameras.

How we built it

We built the frontend using react, vite, and tailwind. We're using supabase as our database to keep track of animal sightings, animal predictions, chat logs, user pings, and more. The bulk of our project consists of a python worker that scans live cameras for animals.

Challenges we ran into

Machine learning model for our Katmai cameras kept on detecting bears as either cows, horses or elephants, and mistaking tree branches for birds. Finding clear and responsive live cameras was difficult.

Accomplishments that we're proud of

Refining the animal classification model we found to increase it's accuracy. Building a working pipeline that captures images from a live stream, sends it through an opencv model, outputs the result to our database, and updates in real-time on the website.

What we learned

We learned the importance of data. If these park cameras were restricted, our project would not have been possible in the first place. Additionally, we had to collect our own data from these cameras to strengthen our model for specific parks.