A web application for tracking and identifying birds at your feeder, built around TensorFlow.js running in-browser against live HLS webcam streams. Designed as a companion app for a smart bird feeder product concept.
- Live bird feeder webcam stream integration via HLS
- Real-time bird detection using CoCo-SSD and MobileNet running entirely in-browser
- Species identification with taxonomy detail — order, family, Latin name
- Sighting log with timestamps
- Personal species catalogue per user
- User authentication via JWT
- Responsive across desktop, tablet, and mobile
React, Redux, Node.js, Express, MariaDB, TensorFlow.js, CoCo-SSD, MobileNet, HLS.js, JWT
Requires Node 14 and MariaDB.
Start the database:
docker run --name weevr-db \
-e MYSQL_DATABASE=birdidapp \
-e MYSQL_USER=birdy \
-e MYSQL_PASSWORD=W33verDB \
-p 3306:3306 \
-d mariadb:10.5Restore the schema:
docker exec -i weevr-db mariadb -u birdy -pW33verDB birdidapp < birdidapp.sqlCreate backend/.env:
DB_HOST=127.0.0.1
DB_USER=birdy
DB_PASS=W33verDB
JWT_SECRET=anystringwilldo
Install and run:
nvm use 14
npm install --legacy-peer-deps
cd backend && npm install && cd ..
npm run devServer runs at http://localhost:3000
Functional as of January 2020. Was self-hosted in production on a Linux server with the React build served statically and the Node backend running separately. Public bird feeder streams that were freely available in 2020 are now largely commercial products — a reasonable validation of the concept. Not actively maintained.