Work / Collaborations
Visual Hand Track
Webcam gesture recognizer: finger-state heuristics over MediaPipe hand landmarks drive two-hand gesture actions.
- Python
- OpenCV
- MediaPipe (via cvzone)
- NumPy
A collaboration with Fluxxion88.
What it does
An OpenCV loop on cvzone’s MediaPipe hand detector that tracks two hands at once. It draws its own 21-landmark skeleton, classifies each finger as up or down from joint-distance heuristics into a five-bit vector, and maps two-hand combinations to actions.
- Fill the quad between both hands’ thumb and index tips.
- Pop a window on a double thumbs-up.
- Exit on a one-hand sign.
How it works
There’s no model training: every gesture comes from geometry on top of the detector’s landmarks, so the whole thing stays one readable loop.
Verified numbers
| Metric | Value | Source |
|---|---|---|
| Detector settings | 2 hands, 0.8 detection confidence | Fluxxion88/visual-hand-track/main.py:11 |
| Main loop | 123 lines | Fluxxion88/visual-hand-track/main.py |