Facial recognition matches a face in a camera image against a stored template or a watch list. Accuracy depends on light, angle, image resolution and the quality of the enrolled photo, and it varies across people. Access decisions from a face alone are risky without a second factor.
Face recognition is the biometric most people have an opinion about, and the one most affected by camera placement. The image is the input, so a poor image produces a poor decision.
How the match is made
Software locates a face, maps it to a numerical template and compares that template with one stored record or many. Verification compares one to one; identification compares one to many. A threshold sets how close a match must be, and lowering it trades missed matches for false ones.
Deployment and privacy questions
Decide whether the system verifies willing users or watches everyone who passes. Consent, signage, template storage and a deletion rule belong in the design. Because accuracy varies with demographics, lighting and angle, a second factor such as a card keeps the door reliable.