Multi-site CCTV control room

The problem

Thirteen construction sites, geographically dispersed, each with its own cameras and its own recorder. Security incidents were being reviewed after the fact, site by site, by whoever could get to a machine. There was no moment at which anyone could see what was happening across the whole operation.

The brief was a single control room. Ninety cameras, thirteen DVRs, one operator, live.

The constraint that shaped everything

Most of the sites are temporary and have no fixed line. They connect over 4G, and mobile operators put subscribers behind carrier-grade NAT.

That single fact removes the standard answer. Behind CGNAT the site does not have a public address of its own, it shares one with a large number of other subscribers, and there is no router you control on the operator side. You cannot forward a port to the DVR because the port you would forward is not yours to forward. Dynamic DNS does not help either, because the problem is not that the address changes, it is that the address is not yours.

Any design that assumed an inbound connection to the site was dead before it started.

The remote access architecture

The answer is to stop trying to reach in. Each DVR establishes an outbound connection to a rendezvous service, the control room does the same, and the two are introduced to each other. Outbound is the one direction CGNAT does not obstruct, because it is what the subscriber connection exists to do.

Built around P2P this way, the deployment gained two properties I would now design for deliberately:

• No inbound attack surface at the sites. There is no forwarded port, so there is nothing on the internet pointed at a construction site DVR. The most common way these systems get compromised is a recorder with its web interface exposed, and that failure mode does not exist here.
• Nothing to reconfigure when a site moves. Construction sites are temporary. A site that relocates or changes SIM keeps working, because nothing in the design depends on where it is.

Making ninety streams fit down 4G

A 4G link at a construction site is not a symmetrical business circuit. Upload is the scarce direction, it is shared, and it degrades with weather, load and the time of day.

The lever is the sub-stream. Cameras publish a full resolution stream for recording and a second, smaller one for live view, and those are independent. Recording quality at the site stays high, because the recording is what an incident review needs. The live stream that crosses the mobile link is tuned down in resolution and bitrate until it fits the worst case rather than the best case.

This is the trade that makes the whole thing viable. An operator watching a wall does not need enough detail to identify a face at forty metres. They need enough to notice that something is happening, and then to open that camera at full quality. Sizing the live path for detection rather than identification is what let ninety cameras coexist on links that could not have carried a fraction of that at full resolution.

The video wall

Eight screens, driven by a workstation built for the job with multiple HDMI outputs rather than an off the shelf machine with a splitter behind it. Splitters duplicate; the wall needed eight genuinely independent surfaces.

The decision I am most pleased with

Ninety cameras across eight screens is not a layout problem, it is an attention problem. Divided evenly it comes to roughly eleven cameras per screen, and at that density a person sees motion and nothing else.

So the layout is tiered by risk rather than by site.

• High-risk zones get space. Entrances and storage areas are shown large, with few cameras per screen. These are where theft and unauthorised access actually occur, and they are worth the pixels.
• Lower-risk areas get density and rotation. General site coverage sits on dense grids that cycle automatically. The operator is not watching these continuously, and pretending otherwise would only dilute the screens that matter.

Weighting the wall by consequence rather than by camera count is the difference between a monitoring centre and a room with a lot of screens in it. One person can hold attention on a small number of high-value views. Nobody can hold attention on ninety.

Result

A single-operator monitoring centre with real-time visibility across thirteen sites, able to respond to an incident as it happens rather than review it the following morning.