Real-time camera analysis with AI vision models 📹✨
Turn your camera feed into interactive scene graphs using local AI. No data leaves your computer.
1. Clone and start:
git clone https://github.com/AmmarMohanna/llm-camera-tracker.git
cd llm-camera-tracker
./run_docker.sh start2. Open your browser:
- Go to http://localhost:8000
- Allow camera access when prompted
- Click "Start" to begin analysis
That's it! 🎉
- Downloads AI model files (~636MB)
- Takes 2-15 minutes depending on internet speed
- Subsequent runs start instantly
- 📹 Analyzes your camera in real-time
- 🧠 Understands what it sees using SmolVLM-500M AI model
- 📊 Creates interactive graphs showing objects, people, and actions
- 💾 Exports data for further analysis
- 🔒 100% private - everything runs locally
- Grant camera permission in your browser
- Click "Start" to begin analysis
- Watch the graph grow as it detects objects and actions
- Adjust frame interval (2s, 6s, 10s) for different speeds
- Export results as JSON when done
- Reset to start over
| Item | Requirement |
|---|---|
| RAM | 2GB minimum |
| Storage | 2GB free space |
| Docker | Any recent version |
| Browser | Chrome/Firefox with camera |
# Start the system
./run_docker.sh start
# Stop everything
./run_docker.sh stop
# View logs if something goes wrong
./run_docker.sh logs
# Get help
./run_docker.sh helpCamera won't work?
- Use
https://orlocalhost(required for camera access) - Check browser permissions
- Close other apps using your camera
Slow or stuck?
- Wait for model download on first run
- Check Docker has enough memory (2GB+)
- Try a different frame interval
Port conflicts?
- Stop other web servers
- Restart Docker:
./run_docker.sh restart
- SmolVLM-500M: Lightweight AI vision model (500MB)
- FastAPI: Backend for processing
- Docker: Easy setup across platforms
- Web Interface: Simple browser-based UI
git pull origin main
./run_docker.sh restartMIT License - see LICENSE file.
Questions? Open an issue • Like it? Star the repo ⭐