Here's a small experiment. Open your CCTV app, pick the camera at your main entrance, and pause on a person walking in. Now zoom into their face.

Can you confidently say who it is? Not "a man in a blue shirt" — the actual person, by name?

For most installations, the honest answer is no. And if you can't identify the face, no AI can either. The information simply isn't in the image.

Megapixels aren't the problem

Owners often assume recognition failed because the camera is cheap. Usually the camera is fine — it's the placement that throws the face away. Three things decide it:

1. The face is too small in the frame

A camera covering a whole compound squeezes a face into a tiny cluster of pixels. From that distance, a face is a smudge — enough to say "someone is there", nowhere near enough to say who. Recognition needs the face to fill a meaningful part of the frame, which means the camera must see people close, not just wide.

2. The camera looks down at heads, not at faces

CCTV is traditionally mounted high — under the roofline, out of reach. Great for theft-proofing the camera; terrible for identity. From a steep angle the camera sees hair, foreheads and caps. A face is recognized when it's seen roughly level, from the front — the way a person at the gate would see it.

3. The light is behind the person

Point a camera at a bright gate from a dark interior and every person entering becomes a black cutout against the glare. The camera exposes for the sunlight, not the face. Backlight is one of the most common reasons an otherwise good camera produces unusable faces all afternoon.

The good news: you don't need 16 recognition cameras

Here's the shift in thinking that saves money: you don't need every camera to recognize faces. You need one or two cameras at the places everyone must pass — the main gate, the entry door, the corridor to the floor. Security people call these choke points.

At a choke point:

  • people walk toward the camera, face first;
  • they pass close, so the face is large in the frame;
  • one well-aimed camera covers everyone, because there's no way around it.

Your remaining wide-angle cameras keep doing what they're good at — watching areas, catching movement, recording context. The choke-point camera answers the question the others can't: who is this?

A 10-minute self-check

For each camera you care about, ask:

  1. Distance — when a person passes, does their face look sharp and large, or like a distant dot?
  2. Angle — is the camera roughly at face height looking forward, or high up looking at scalps?
  3. Light — is the person's face lit, or is the bright light behind them?

Any camera that fails two of the three will not reliably identify people — no matter what software you put behind it. The fix is rarely a new system; it's usually repositioning one camera, or adding a single camera at the right doorway.

Recognition isn't bought with megapixels. It's earned with placement.

Not sure which of your cameras make the cut? Get a free per-camera readiness score — we grade each feed for placement, angle and light, and tell you the one change worth making.