August 28, 2026

Face authentication in real conditions: what the demo never shows

Every vendor demo works. Shift change in the rain, with masks and backlight, is the test that matters.

Face authentication is now accurate enough that accuracy is rarely the deciding factor. Throughput, lighting tolerance and spoof resistance are.

Measure at the queue, not the door

A 0.9 second match sounds fast until 300 people arrive in eight minutes. At that volume the difference between 0.4 s and 0.9 s is the difference between a flow and a queue that ends with the door propped.

Ask for the lux range

Glass lobbies hit 25,000 lux in direct sun. Loading bays sit near zero. A single stated accuracy figure means nothing without the illumination range it was measured across.

Insist on ISO/IEC 30107-3

Presentation attack detection should be tested against print, screen, mask and 3D replica attacks by a third party. “Liveness detection” as a bullet point is not a claim you can verify.

Need more information?

Tell us about the site, the doors and the standards you have to meet. A Xipheredge specialist will come back within one business day — no obligation.