Built a gesture-controlled calculator a while back using MediaPipe. Extracted the detection logic into a standalone library so anyone can add gesture recognition to their project without touching CV code.
from mp_gesture_lib import GestureDetector
detector = GestureDetector() # bundled model, zero config
result = detector.detect(frame) # pass any BGR webcam frame
print(result.gesture, result.confidence)
What it detects out of the box:
- Finger count 1–10 (geometry-based, no ML)
- Math ops: plus, minus, multiply, divide, equal, clear (ML model, bundled)
- Two-hand rules for plus/multiply (landmark geometry)
- Returns
"unknown"cleanly when nothing matches
Custom model support — drop your own .task file, it's checked first. Bundled model is fallback. Any label passes through raw, no hard-coded mapping.
pip install mp-gesture-lib
📖 Docs: debabratasaha-dev.github.io/mp-gesture-lib-package
🐙 GitHub: github.com/debabratasaha-dev/mp-gesture-lib-package
Feedback welcome — especially on the gesture pipeline priority logic. If you find it useful, I’d really appreciate a ⭐️ on GitHub!
Top comments (0)