← All work
Personal project · Computer vision
Facet
Real-time facial geometry analysis from a webcam feed.
- Year
- 2025
- Stack
- Python, Flask, MediaPipe, OpenCV, NumPy
- Links
- Source
The problem
Facial proportions are usually judged by eye, which makes them subjective and hard to reproduce. I wanted objective measurements, computed live from an ordinary webcam.
What I built
A Python and Flask service that tracks 468 facial landmarks per frame with MediaPipe and OpenCV. A NumPy geometry engine turns those points into five proportion metrics and streams them to a browser dashboard.
A decision worth explaining
Score a ten-second window, not a single frame
Per-frame measurements shift with head pose and lighting, so the app averages every metric across a ten-second scan before locking a final score. It trades instant feedback for results that repeat from one run to the next.
Outcomes
- 468 landmarks per frame at under 50 ms per frame (24 fps).