AI worker (YOLO11 + ByteTrack + OCR) -> Cloudinary (authenticated storage + metadata + face blur on delivery) -> FastAPI + DB -> live map dashboard.
cp .env.example .env # add your CLOUDINARY_URL
cd backend && pip install -r requirements.txt
export $(grep -v '^#' ../.env | xargs) && uvicorn main:app --reload
# open http://localhost:8000 and click "Load demo data"Worker (second terminal, a GPU is not required for yolo11n):
cd worker && pip install -r requirements.txt
export $(grep -v '^#' ../.env | xargs)
python worker.py --source traffic.mp4 --camera-id cam01 --name "MG Road" --lat 12.9756 --lng 77.6066
python worker.py --source traffic2.mp4 --camera-id cam02 --name "Trinity Circle" --lat 12.9730 --lng 77.6200Run it once per "camera" with different clips/coordinates, then search the plate in the dashboard.
- Cloudinary: create a free account, copy the API environment variable (
CLOUDINARY_URL) from the dashboard. - Database: create a free Supabase (or Neon) project and copy its Postgres connection string as
DATABASE_URL. - GitHub: push this folder.
- Render: New > Web Service > connect repo > Root Directory
backend> Runtime Docker. Environment variables:CLOUDINARY_URL,DATABASE_URL,WORKER_KEY(long random string), optionalTELEGRAM_BOT_TOKENandTELEGRAM_CHAT_ID. - Open your Render URL, click "Load demo data" to check it works.
- Run the worker on your laptop pointing at the live API:
export API_URL=https://YOUR-APP.onrender.com WORKER_KEY=... CLOUDINARY_URL=...then thepython worker.py ...commands above.
Download an Indian number-plate dataset (Roboflow Universe/Kaggle, YOLO format), train in Colab with yolo train model=yolo11n.pt data=data.yaml epochs=50, and save the result as worker/plate.pt. The worker uses it automatically.