Zermatt/Research/Photometric Stereo
Jump to navigation
Jump to search
Photometric Stereo for Ear Scanning[edit | edit source]
May 7, 2026
Technique[edit | edit source]
Single fixed camera captures images illuminated sequentially from different LED positions around a ring. Surface normals and 3D depth reconstructed computationally. Reported accuracy: 0.05mm. Introduced by Wolfgang Herwald, April 28, 2026.
MVPS Studio (Reference Implementation)[edit | edit source]
- Developer: Dr. Berkan Lafci, ETH Zurich | Published: February 2024
- GitHub: https://github.com/berkanlafci/mvpsstudio
- Hardware: 24-LED NeoPixel ring (~66mm OD), Raspberry Pi 5, Ubuntu 22.04, GPIO pin 10
- Software: Open-source deep learning reconstruction pipeline for surface normal estimation
Optics[edit | edit source]
- Diffuser: Rosco Cinegel 3026 Tough White Diffusion (=LEE 216), 20x24" sheet — 36% transmission (Y), 1.5 stop loss. Cut to fit over LED ring to soften point sources.
- Polarizers: Selens 7.8x11.8" adhesive linear polarizing sheets (x2) — cross-polarization eliminates specular glare from skin. One sheet over LED ring (source), one over camera lens (analyzer), rotated 90° to each other.
Calibration Spec[edit | edit source]
LEE 216 = Rosco Cinegel 3026 = 36% luminance transmission, 1.5 stop reduction. Use this when accounting for diffuser attenuation in photometric stereo light intensity calculations.
Controller[edit | edit source]
Raspberry Pi 5 (8GB RAM), Ubuntu 22.04. GPIO pin 10 (physical pin 19) for LED data line.