Understanding Face and Eye Visibility in Front-Facing Cameras of Smartphones used in the Wild
summary
Field study installs a recorder app on 11 phones for two weeks and collects 25,726 in‑the‑wild photos plus sensor and app logs; full face appears only 29% full face visible, both eyes in ~48%, and ~30% show no face. OpenFace finds faces in 75.9% of images with a whole face but only 33% overall. Failures stem from motion blur, lighting, occlusion and posture. Recommendations: use sensors to trigger processing, build detectors for partial faces and add explicit privacy signals.
AI-written summary · about
chapters (7)
- 0:00Introduction and motivation
- 3:18Field study methodology
- 7:12Face visibility findings
- 9:16Algorithm performance analysis
- 10:43Reasons for poor visibility
- 12:31Privacy implications
- 14:42Summary and future directions
related talks
06mFacial Recognition: Understanding Privacy Concerns and Attitudes Across Increasingly
26mVirtual U: Defeating Face Liveness Detection by Building Virtual Models...
40mNew Face, Who Dis? Protecting Privacy in a World of Surveillance
01h 05mFacial recognition, a 'convenient' and 'efficient' solution looking for a problem?
37mSeeing is Believing: Mobile Vision API Deep Dive
30mMobile applications that can see
06mInSight: Monitoring the state of the driver in low-light using smartphones
31mFace Recognition