Face recognition camera that sends events, not streams.

What the camera actually does

The body detects a face in the frame, extracts it, and compares it against the list held on the device. A match becomes an event with a snapshot, a time and a name from the list, and that event is what the recorder or the VMS receives.

The list size, the number of faces per second and the recognition distance are stated per model in the data sheet, not on this page. Ask for the sheet for the model you are testing.

Light and angle decide the result

Recognition needs the face lit and roughly frontal. A camera over a door at 2.4 m facing north reads most people who walk in; the same camera facing a glass door at noon reads very few.

Test with your own footage before you promise a rate. A figure measured at a controlled door does not transfer to a wet loading bay with people in hoods.

What it means for the system

The event stream is small, so a 4 channel recorder can hold a door group that would otherwise need a server. Bandwidth and disk both drop because the camera sends a face crop rather than a 4MP stream.

Plan the list management before installation. Someone has to add, remove and audit the face list, and that is a process question rather than a camera question.