A person on camera isn't an event. It's a question.

Every automated CCTV system in the world is built to answer one question: is there a person in the frame? Point a modern camera at a scene and it will tell you, reliably, when a human appears. That problem is solved, and has been for years.

The trouble is that "is there a person" is almost never the question that matters.

Consider a single, ordinary situation. It's 1am on a lightly-manned site, and a figure appears on camera walking across the yard. Presence detection does its job and fires an alert. But whether that alert means anything depends entirely on things the camera can't see in the frame in front of it. Is that the security guard, doing exactly what they're paid to do? A contractor who's meant to be there? Or someone who shouldn't be anywhere near the place? Same pixels. Same "person detected." Three completely different events, and only one of them is worth waking someone for.

A system that can only detect presence has no way to tell them apart. So it does the only thing it can: it alerts on all three, and leaves a human to sort out which is which.

Where this breaks in practice

The clearest example we've come across is one that security teams will recognise immediately. On a site with a guard and cameras, what happens when the guard goes on patrol?

Every camera they walk past detects a person and fires. Person on camera three. Person on camera seven. Person on camera twelve. It is, of course, the guard — but the system has no concept of "the guard", so each camera reports it as a fresh event, and the control room is buried in alerts that are all the same harmless thing. The rational response, the one that happens on real sites every night, is to turn the cameras off for the duration of the patrol.

Think about what that means. At the exact moment there is definitely a person moving around the site, the cameras get switched off — so if anything happens anywhere else while the guard is walking the far fence, nothing is watching. The system's inability to understand one simple piece of context (this person is the guard, and that's fine) forces a choice between drowning in false alerts and going blind. Neither is security.

This is not a failing of the cameras. The cameras see everything perfectly. It's a failing of a model that stops at detection and never reaches understanding.

What understanding actually looks like

The fix isn't a better presence detector. It's a system that holds context — that knows the guard is on patrol, expects to see them, and treats seeing them as the non-event it is, while continuing to watch everything else with full attention.

That's what we built patrol mode to do. When a guard goes out, the system understands they're out. It stops firing alerts every time it sees them, because it knows what it's looking at. But it keeps watching the rest of the site, so if something happens on the north perimeter while the guard is checking the south gate, it tells them — turning the guard from someone who had to blind the cameras to move into someone the cameras are actively helping.

It's a small feature. But it's a clean illustration of the whole difference. Presence detection would see the guard as an alarm. Understanding sees the guard as the guard — and keeps its attention on what actually matters.

The general point

Patrol mode is one case, but the principle underneath it is the entire game. The value in security monitoring was never in noticing that something moved. It's in knowing what that movement means: threat or routine, stranger or staff, the ten seconds that matter or the ten thousand that don't. A person on camera is not an event. It's a question — and the only systems worth having are the ones that can answer it.

If that's a conversation worth having, Volpex is built to be exactly that layer. Get in touch.