logo

Manufacturing / Access and identity

Face recognition in factories, and when not to use it

Face recognition identifies an enrolled person from their face. In factories it is frequently the wrong choice, because the equipment mandated in most zones covers the face, and because employee facial data carries obligations that the underlying task rarely requires. It works at the administrative entrance and poorly beyond it.

Reviewed by Dr. Vivek Gujar, Chief Strategy Officer  ·  Updated 31 August 2026  ·  7 min read

The conflict nobody mentions in the demo

Face recognition needs an unobstructed view of the face. Industrial PPE is designed to cover it. Helmets shade the brow, safety glasses obscure the eye region, face shields and welding masks block it entirely, and dust masks cover the lower face. The zones where you most want to know who is present are precisely the zones where PPE is mandated, which means facial recognition is least reliable exactly where it would be most useful.

This is not a model quality problem that improves with a better vendor. It is geometric. A worker in compliant PPE on a production line may have very little face visible, and a system that recognises them anyway is either seeing a non-compliant worker or is matching on something other than the face.

The practical consequence is that face recognition in manufacturing works in a narrow band: the administrative entrance, the office block, and the gate before workers don equipment. Which is a real use case, and a much smaller one than most proposals assume.

Match the tool to the actual question

What you are trying to know, and what answers it with the fewest obligations
QuestionAppropriate toolWhy
Is this person authorised to enter?Card or credential, optionally with face at the office entranceA credential answers it directly, works with PPE on, and creates no biometric record.
How many people are in this zone?Person detectionCounting needs no identity at all.
Where did this person go across the site?Re-identificationAppearance matching answers path questions without determining who anyone is.
Is everyone in this zone wearing required PPE?PPE detectionThe compliance question is about equipment, not identity.
Who exactly was present during this incident?Access control records plus footage reviewThe credential log already answers it, with a clear evidentiary trail.
Did this specific person clock in?Attendance system at the entranceA single controlled point, not site-wide recognition.

Most requests that arrive as "we want face recognition" are answered by one of the other rows. Working through the actual question first usually removes the requirement, along with the obligations it would have brought.

The obligations you take on

Facial data is a distinct category. Deploying it against employees means holding an enrolled gallery linking faces to identities, which brings duties that a card reader does not:

Rule 4 commences on 13 November 2026, with Section 33 penalty provisions following on 13 May 2027. Any facial recognition deployment involving employees should be reviewed by your own counsel before commitment. This page is general information, not legal advice.

Where it is defensible

None of the above makes facial recognition wrong everywhere. It is defensible where the case is narrow and the alternatives genuinely fail:

In each case the population is small, the location is a controlled single point, and PPE is not in the way. Site-wide recognition across a production floor meets none of those conditions.

If you deploy it anyway

Keep the gallery and the matching on site through edge processing, so facial templates never leave the plant. Restrict the gallery to the specific population and specific doors that require it. Set a retention period for match logs tied to a stated purpose. Join enrolment and deletion to the HR joiner and leaver process. And document the alternative available to anyone who declines, before the system goes live rather than in response to the first complaint.

Frequently asked questions

Does face recognition work with helmets and safety glasses?

Poorly. Helmets shade the brow, safety glasses obscure the eye region, and face shields or dust masks block the face entirely. This is geometric rather than a model limitation, so a better vendor does not fix it. Facial recognition works reliably in factories only where PPE is not worn, such as the administrative entrance.

What should we use instead for shop floor access?

A card or credential answers the authorisation question directly, works with PPE on, and creates no biometric record. Where the question is about paths or counts rather than identity, person detection and re-identification answer it without any enrolled gallery. Start from what you actually need to know, not from the technology.

Can we require employees to enrol in facial recognition?

Consent obtained from an employee by an employer is a contested basis, because refusal carries consequence. If enrolment is effectively mandatory, calling it consent is hard to sustain. Design a genuine alternative route to work for anyone who declines, before deployment. Have your own counsel review any employee facial deployment.

Is face recognition ever the right choice in a plant?

Yes, in narrow cases: the administrative entrance for office staff and visitors, high-security areas with a small enrolled population, and contractor verification at the gate before PPE is donned. What these share is a controlled single point, a limited population and no PPE in the way. Site-wide floor recognition meets none of those.

What happens to facial data when an employee leaves?

It has to be deleted, and that needs to be joined to your HR offboarding process rather than left to a periodic cleanup. Unlike a card, a face cannot be reissued if a gallery is breached, so retaining enrolment records for former employees carries risk with no operational benefit.

Deciding between face, card and appearance matching?

IndoAI assesses what your access and tracking questions actually require, and frequently recommends against facial recognition where a credential or appearance matching answers the same question with fewer obligations.

Talk to an adviser

Reviewed by Dr. Vivek Gujar, Chief Strategy Officer at IndoAI Technologies Pvt. Ltd., a Pune-based edge AI camera platform founded in 2021. He reviews IndoAI's published technical claims for accuracy. Profile