Real-time lip-parting detection using MediaPipe face landmarks and OpenCV. Creates a GUI with realtime lip detection and live-threshold control via slider. Writes out a .csv with timestamped parting events. Built for detecting silent speech miming via webcam.
The script tracks two inner-lip landmarks (indices 13 & 14) from MediaPipe's face landmarker and computes the normalised vertical gap between them. When the gap exceeds a threshold, a "MIMING" flag is shown on-screen. Events (start, end, duration) are logged and exported to CSV.
The threshold is adjustable live via a GUI slider — useful because the optimal value varies between subjects (e.g., facial hair and camera angle affect it).
GUI with live threshold slider and gap readout:
Lip landmark overlay — lips closed (below threshold):
Lips parted — detection triggered:
python MediaPipe_LipParting_QuickDemo.pyDependencies (mediapipe, opencv-python) and the face landmark model are downloaded automatically on first run. Press Q to quit the GUI.
When you quit the GUI (press Q), all detected lip-parting events are written to a .csv file with four columns:
- event — sequential event number
- start_s — event start time (seconds since script launch)
- end_s — event end time
- duration_s — how long the lips were parted (milliseconds)
This makes it easy to analyse miming patterns, count events, or feed the data into downstream processing.
All tuneable parameters are at the top of the script in the "Master":
| Parameter | Default | Description |
|---|---|---|
LIP_OPEN_THRESHOLD |
0.004 | Starting threshold (adjustable live via slider) |
EXPORT_CSV |
True | Toggle CSV event export |
CSV_FILENAME |
lipparty_events.csv |
Output filename |
FLAGMESSAGE |
MIMING |
On-screen alert text |