Put a person in front of a wall of forty monitors and, within a few minutes, they are not watching forty cameras. They are watching none of them. Attention does not divide that way. The guard is not failing at the job; the job as designed cannot be done by a human, and pretending otherwise means the cameras are really only useful afterwards, when something has already happened and somebody goes looking for the tape.
The fix is to turn the arrangement around. The system watches every camera, all the time, and the person decides what to do about the handful of things it finds.
The guard’s attention is the scarcest thing on site
A security team’s most valuable resource is a trained person’s judgement, and a monitor wall is the most expensive possible way to waste it. Hours of it go on footage in which nothing happens, and the moment something does happen it is as likely as not to be on the screen they were not looking at.
What a person is actually good for is responding: walking to the gate, making the call, deciding whether the thing at the fence is a problem. A system that hands them a short list of things worth a decision, with the footage attached, gives them their job back.
Search a person across every camera
This is the capability that changes how a site is run. Someone has been seen at the east gate; where else have they been today? With a monitor wall, that question is an afternoon of scrubbing through cameras one at a time. With the system watching, it is a query: the same person is recognised across every camera on the estate, tracked from zone to zone, and their movement through the site comes back as one path.
It works on people whose faces are covered, which at most sites is a lot of people. It works on the footage you have already recorded, so last week’s question can be answered as easily as today’s. And because the same layers read number plates and detect vehicles and drones, the question can be about a car or an aircraft as easily as a person.
The alert that matters most
Most security incidents do not arrive without warning. They are preceded by someone looking. Reconnaissance looks ordinary from a single camera on a single day, which is exactly why a person watching monitors never sees it. Seen across every camera and across the week, it is a pattern, and the system raises it as an alert with the footage attached:
- Someone lingering at the perimeter — not passing, staying — near the fence, the gate or the loading bay, for longer than anyone with a reason to be there would.
- The same unknown person back again, at the wire on Monday and again on Wednesday, when nobody on Wednesday’s shift was there on Monday.
- A person who was seen outside a couple of days ago now inside the premises. This is the one that matters most. Somebody who studied the site from the outside and then came in is the shape that a great many bad outcomes take, and it is invisible to a guard because the two clips are days apart and on different screens.
Each of these is a notification with the clips behind it, before there is an incident to review. A person then decides whether it is the contractor who forgot their pass or something else, which is the decision people are good at and monitors are not.
It works with the cameras you have
None of this needs a new camera estate. The system sits over the one that exists — CCTV, body cameras, dashcams — in the formats those cameras already produce. Where the footage is too dark or too noisy to be useful, the enhancement toolkit underneath makes it readable, and that toolkit runs on a single GPU laptop rather than a server room.
It runs on site, with no external calls, which is the normal configuration for this kind of system. We publish no recognition-accuracy figures; we run it against your own footage instead, because your cameras and your lighting are the only benchmark that counts.
The cameras were always capable of watching everything. Now something is.