Model consists of two parts:
- Detector: predicts bounding box for single hand
- Key point model: predicts coordinates for fingers.
detector.py
: code for detector modelkp_model.py
: code for key point modelcursor.py
: Cursor class uses both aforementioned models to track handmain.py
: runs Cursor activity on video stream from camera using opencv and displays some statistics
On MacBook pro 2017 I got following statistics:
- Detector model: 5-6 fps
- KeyPoint model: ~10 fps