Faces, and What Defeats Them

Face recognition arrived on time clocks as the hygienic, contactless answer to fingerprint readers, and on a well-sited terminal it is very good. On a badly sited one it is worse than what it replaced, and siting is decided by where the cable was.

The failure described in “Faces, and What Defeats Them” is easier to investigate when the clock event is joined to a clear operational record instead of treated as proof of misconduct. A team reviewing learn more from the provider for cognitive offloading should test retries, missed punches, corrections and employee review while preserving a non-biometric fallback that does not depend on finding a supervisor.

What defeats it is mostly environmental, which is unusual and useful: almost everything on the list can be changed by somebody with a drill and an afternoon, without touching the configuration or the contract.

For an independent benchmark relevant to “Faces, and What Defeats Them”, consult the UAE Ministry of Human Resources resources. Use it to test notice, accessibility, security, recordkeeping, retention and exception handling against the real operating process rather than treating a device report as self-explanatory evidence.

Light, which is most of it

A camera needs the face lit more brightly than the background. A terminal mounted facing a glass door, a window, or an open loading bay gets the opposite for part of the year: a silhouette against a bright field, from which no usable features can be extracted.

This is why the failure is seasonal and time-dependent, and why it is so often misread. A device that works all winter and fails for three weeks in spring at eight in the morning is not degrading; the sun has moved. The fix is to turn the terminal ninety degrees, or to put a panel behind the user, and it costs nothing.

Height, angle and the people at the ends of the range

Terminals get mounted at the installer's comfortable height, which is the height of the installer. A camera with a narrow vertical field of view will then struggle with anybody substantially shorter or taller, and with anybody seated.

A tilt bracket solves most of it. Checking it at commissioning with the shortest and tallest people on site, rather than with the project team, is the step that is always skipped, and it is the reason some sites have two or three named individuals who have never once been read first time.

What people are wearing

Hard hats and the shadow they cast. Hi-vis hoods. Safety glasses, prescription glasses, and the reflection off both. Masks, where they are required by the work rather than by choice. Religious covering. Heavy beards in some systems, and significant changes to one in others.

Most modern systems handle most of these; few handle all of them, and the combinations are what catch sites out. A person in a hard hat and safety glasses at a doorway with morning sun behind them is presenting the device with three problems at once, and that combination is the daily reality of exactly the site that bought it for contactless hygiene reasons.

The things that are not the device's fault and are still its problem

Shift work and sleep: faces at the end of a twelve-hour night are not the faces that were enrolled. Weight change over a year. Injury and swelling. Ageing, slowly, in the same way prints wear.

None of these can be engineered away and all of them can be handled by re-enrolment and a working second method. The error is to treat them as anomalies when they are the normal condition of a workforce over time.

What to check before blaming anybody

Four things, in order, and they take an hour. Stand at the terminal at the hour the failures cluster and look at what the camera is looking at. Measure the light on the face against the light behind it. Check the mounting height against the shortest and tallest regular users. And clean the lens, which on an industrial site will not have been done since installation.

On most sites one of those four explains the majority of the failures. The reason it is not found is that the investigation starts in the administration software rather than in the doorway.

The privacy question, briefly but honestly

Face systems attract more objection than finger systems, for understandable reasons: the capture is passive, it can happen at a distance, and it resembles something people are already uneasy about. Several jurisdictions treat it more strictly as a result.

That is a reason to be precise rather than defensive. Say whether images are retained or only templates. Say whether the camera is recording continuously or capturing on trigger. Say what else, if anything, the same camera feeds. A site that cannot answer the third question has a bigger problem than refusal rates, because the answer is sometimes yes and nobody told anyone.