In many ways, the deployment model described in the linked article is the perfect facial recognition app.
It’s minimalist, requiring no large database of potentially sensitive information. If, in the case of the most minimal possible system, the database just contains faces it’s debatable whether or not the face alone is personally identifiable information (PII) at all which ought to alleviate the concerns of even the most strident privacy advocates. The database can be wiped clean altogether every few hours.
Since matching algorithm developers tend to charge by the size of the searchable database, the small database size should keep the system affordable.
The environment in which the enrollment image is created can be tightly controlled.
The probe image is collected on the same day as the database image, so no one is likely to age by several years, have plastic surgery or grow a beard in the time between photos.
The environment where the probe image is created can also be tightly controlled.
The people to be identified are cooperative users of the system, so it’s more like an access control deployment and less like surveillance.
Databases like these are likely to remain very small making error rates manageable by an attendant human being.
See also: One-Time-Only ID Technologies
Automated face recognition speeds up plane boarding (The Engineer)
In use, passengers reaching the ‘self-boarding’ gate pass through an automatic electronic barrier which takes an infrared scan of their face.
This information is checked against the biometric data that was taken at the check-in stage.
When the two sets of data scans are successfully matched, the barrier opens and the passenger can pass through and board their flight.