Facial recognition identifies or classifies a person from an image of their face. In signage it is usually pitched for audience measurement: estimating age and gender to report who saw a campaign, or to change what plays.
The critical distinction, which vendors blur constantly, is between detection and recognition. Detection counts that a face is present and possibly estimates a bracket, discarding the frame immediately. Recognition creates a signature that can match the same person again, across visits or across sites. They look identical in a demo and are entirely different objects in law.
The caveat: recognition processes biometric data, which GDPR and Turkey's KVKK both treat as a special category needing an explicit lawful basis, usually consent you cannot realistically obtain from someone walking past a screen. "We do not store images" is not a defence; the template derived from the face is itself biometric data. Several European regulators have fined retailers for exactly this deployment.
The practical position: anonymous detection with on-device aggregation and no persistent identifier is defensible and usually sufficient for the business question. If a proposal needs recognition, it needs a lawyer before it needs a budget.