Air Writing

Real-time hand tracking that turns finger motions into a virtual pen. Draw digits and simple math expressions in the air, let the app recognize what you wrote, and see the result on screen.

How it works

1. MediaPipe tracks your hand and detects when only the index finger is up.
2. The index fingertip path is drawn onto an invisible canvas.
3. On save, the canvas is preprocessed and segmented into characters.
4. A trained TensorFlow model predicts each character.
5. The predicted expression is evaluated and the answer is shown.

Repo layout

• src/air_writing/finger_track.py - main application
• models/air_writing_model3.keras - trained model used by the app
• notebooks/ - experiments and training notes
• scripts/ - helper/legacy scripts
• outputs/ - saved drawings

Setup

pip install -r requirements.txt

Run

python src/air_writing/finger_track.py

Controls

• c clear canvas
• n lift pen (start a new stroke)
• s save drawing + evaluate
• q quit