Business CCTV: How to Choose a System That Actually Produces Evidence

The cameras went in. Then something actually happened, you opened the recording, and what you found was a blurred figure whose face is no use in a police report or an insurance claim. That is usually not a "camera quality" problem. It is the result of nobody deciding, in advance, exactly what each camera was supposed to see.

Start with one question: what do you want to see?

There is a wide gap between knowing that somebody entered the stockroom at two in the morning and proving that it was one particular person. The international standard IEC 62676-4 reduces that gap to a single measurable number: pixel density, the number of pixels falling on each metre of scene width at the target's distance. Its four levels are known by the acronym DORI.

• Detect — 25 px/m: there is a person or a vehicle. Nothing more.
• Observe — 62 px/m: you can follow movement and make out the colour of clothing.
• Recognise — 125 px/m: you can place somebody you already know, such as one of your own staff.
• Identify — 250 px/m: you can make out a stranger's features to a standard that will serve as evidence.

The arithmetic is simple: divide the camera's horizontal pixel count by the scene width in metres. A 4MP camera (2560 pixels wide) covering an aisle 10 metres across gives you around 256 px/m, which is identification level. The same camera on a scene 20 metres wide gives you 128 px/m: enough to recognise a member of staff you know, not enough to make out a stranger's features. The camera has not changed — the scene got wider.

Why "30m night vision" does not mean you will see a face at 30m

The night vision figure in the datasheet describes how far the infrared illuminators reach, that is, how far the light travels — not how far away you can recognise a face. The two numbers have no direct relationship.

Take the most common configuration on the market: a 4MP camera with a fixed 2.8mm lens on a 1/2.8-inch sensor. Its horizontal field of view works out geometrically at close to 90 degrees, and many datasheets quote a wider figure, above 100 degrees. At 90 degrees, the scene at a distance of 30 metres is about 60 metres wide: divide 2560 by 60 and you get around 43 pixels per metre — and at 100 degrees the density drops to about 36. Either way you are barely above the "detect" threshold and a long way short of "recognise": the camera tells you somebody is there, and that is all, however powerful the night-time illumination.

To reach identification level at 30 metres with the same camera, you need a scene no more than about ten metres wide — a lens in the region of 16mm with a narrow field of view of roughly 19 degrees. That is where the real trade-off sits: a wide lens covers area and identifies nobody; a narrow lens identifies features and covers one doorway. The answer is not to buy a "more powerful" camera but to split the job between cameras: a wide camera on the general scene to tell the story, and a narrow one fixed on a choke point everyone has to pass through to capture the face.

The sensor and the lens come before the megapixel count

More megapixels on the same sensor size means smaller pixels, and a smaller pixel gathers less light. An 8MP camera on a 1/2.8-inch sensor can be worse at night than a 4MP camera on the same sensor: more detail by day, more noise after dark. A larger sensor and a wider aperture (a lower f-number) do more for low light than any extra megapixels.

The item usually overlooked is WDR, wide dynamic range. A camera facing a glass door in daylight will turn everyone who walks in into a black silhouette against the street light, and raising the resolution will not fix it. What does fix it is true WDR built on combining multiple exposures (not the "digital" kind, which merely processes the image) in the region of 120dB, together with one simple installation rule: do not have the camera facing the light source head-on — set it at an angle to the entry path.

A varifocal lens costs more, but it lets you adjust the field of view on site after installation. With a fixed lens, a mistake in the focal length can only be corrected by replacing the camera itself.

IP or analogue?

Modern analogue — HD over Coax in its AHD, TVI and CVI variants — with a DVR is still a reasonable choice in one case: you already have coax runs in place and working, and you want to upgrade the cameras without breaking the walls open again. It costs less, and it tolerates runs well beyond a hundred metres without intermediate equipment, but it usually needs a separate power cable to each camera, and its options for analytics and high resolution are narrower.

IP systems with an NVR are the default choice for any new installation: a single network cable carries data and power together, resolution is higher, analytics run on the camera itself (line crossing, intrusion into a zone, telling a person from a vehicle to cut false alarms), and the cameras can be isolated on a subnet of their own. Its main constraint is that a copper network run is limited to a hundred metres.

Cabling and PoE — where most of the problems are buried

Most faults blamed on the cameras come from the cabling or the power. The points that deserve your attention:

• Distance: 100 metres maximum for a copper run (about 90 metres of fixed cable and 10 metres of patch leads). Beyond that you need fibre with a media converter, or a PoE extender — and the extender draws on the power budget and adds a point of failure.
• PoE standard: 802.3af supplies about 15.4W from the switch, 802.3at (PoE+) about 30W, and 802.3bt reaches 60W or 90W depending on the class. An ordinary fixed camera is fine on af; one fitted with a heater against condensation, or a PTZ camera, needs at or above.
• The switch's total budget: this is the best-known mistake. An eight-port PoE switch may have a total budget of only 60W or 65W — meaning it cannot feed eight cameras drawing 10W each. Look at the total budget, not the capacity of a single port.
• Outdoors: UV-resistant cable inside conduit, sealed junction boxes, and earthing and surge protection for cameras mounted on roofs and perimeter walls.
• Numbering and documentation: every cable numbered at both ends, and a schedule showing which port serves which camera.

That last point looks like a formality until the first fault. On the branch and head office network project for Al Raya Market, numbering the cable ends and dressing them inside the racks was an integral part of the handover, and the same logic applies to camera cabling.

Working out storage: the number that surprises everybody

The formula you need: every 1 Mbps consumes about 0.45GB per hour, or roughly 10.8GB per day.

A worked example: eight 4MP cameras encoding in H.265 at 3 Mbps each, recording continuously, with 30 days' retention. The sum: 8 × 3 × 10.8 = about 259GB a day, and multiplied by 30 that reaches roughly 7.8TB. Drop the bitrate to 2 Mbps and the figure falls to about 5.2TB; raise retention to 90 days and it climbs to about 23TB. With the older H.264 codec, expect higher consumption at comparable quality: somewhere between one and a half and two times that of H.265, depending on the scene.

Three things will spoil that calculation if they are ignored:

• Bitrate rises at night: low-light noise looks to the encoder like constantly changing detail, so consumption goes up under variable bitrate recording (VBR). A scene busy with movement does the same, so size for the worst case rather than a comfortable average.
• The drives: CCTV writes around the clock without pause, a different workload from a desktop drive. Use surveillance-rated drives, and work out the usable capacity after RAID if you are using it.
• The NVR's limits: the channel count is not everything. Look at the ceiling on incoming data rate in Mbps, the number of drive bays, H.265 support, and the number of built-in PoE ports and their budget where they exist.

The retention period itself is set by the answer to one question: how many days does it usually take before you discover a problem? If stock shortfalls only surface at the monthly count, a week of retention is worth nothing in practice.

Where the cameras actually go

In a retail shop, the priority is not "covering every corner" but securing specific points:

• The entrance: a relatively narrow camera at a height of 2.2 to 2.5 metres, facing the entry path to capture the face. Mounting it four metres up "to protect it from tampering" is the quickest route to footage of the tops of people's heads and nothing else.
• The checkout: two cameras, in fact — one overhead that sees the cash drawer and the movement of hands, and another that sees the customer's face and the cashier's.
• The stockroom door and the rear delivery door: the two points where goods really leave.
• Aisles holding high-value stock and the corners staff cannot see from where they stand.

In a warehouse or a factory the logic is different: the loading and unloading bays, the picking and packing bench (where most quantity discrepancies arise), the ends of the aisles rather than their middles, and the building perimeter and the gate. The governing rule in both cases: every camera has a written job, and if you cannot put that job into one sentence, you are paying to store footage that earns you nothing.

The legal side in Egypt: recording staff and customers

What follows is a practical explanation, not legal advice; how the law applies to your own case is a matter for a legal adviser.

Camera footage in which people can be identified is personal data within the meaning of Law 151 of 2020. The law's executive regulations were issued by Decision 816 of 2025 in November 2025, and with them a one-year grace period for bringing operations into line, ending on 31 October 2026 — a few weeks after this article is published, not a year away.

What matters most for cameras in the regulations: a licence from the Personal Data Protection Centre to operate video surveillance systems in public places, with private homes excepted. The official fees are modest, but the real issue is not the fee — it is the classification: does your shop, your warehouse or the entrance to your premises count as a "public place" within the meaning of the regulations? That is a question for a lawyer before the installation, not after it. The regulations also require a sign informing those present that recording is taking place.

In practical terms, what you can do from today:

• A clear sign stating that recording is in progress, at every entrance — this is a regulatory requirement, not merely good manners.
• A defined, written retention period — keeping footage forever is not a neutral choice.
• Restrict who can view the recordings: a named account for each person, not a shared admin account, with viewing and export operations logged.
• Tell staff in writing that the monitoring exists and what it is for, and include it in the workplace rules. And since Labour Law 14 of 2025 replaced Law 12 of 2003 with effect from 1 September 2025, reviewing your internal rules against it is a good opportunity to document this once and for all.
• Absolutely off limits: toilets, changing rooms, prayer rooms and private rest areas. This is not a question of taste: Article 309 bis of the Penal Code makes it an offence to take a picture of a person in a private place without their consent.

We covered the law's operational effect on infrastructure more broadly in a separate article on Data Protection Law 151 and what it means for your infrastructure.

Do not leave the cameras themselves as a way in

A camera and an NVR are computers attached to your network. Three mistakes recur: leaving the default password in place, opening the NVR's ports straight onto the internet through port forwarding so the manager can watch from his phone, and leaving the cameras on the same company network that carries the accounting systems.

The alternative: a separate VLAN for the cameras, external access over a VPN or the manufacturer's own app with strong passwords rather than opened ports, and firmware updated on a schedule. This part in particular belongs to the network infrastructure and the firewall, not to the camera — and it is the same logic the FortiGate and Sophos appliances we supply and install are built on.

What actually determines the cost

The cost of any camera system moves with clear line items rather than being a fixed figure per camera: the sensor class, the lens and whether there is WDR; the channel count and the NVR's ceiling; storage capacity worked out against a realistic retention period; cable lengths and the nature of the route (suspended ceiling, or conduit through existing walls); the PoE switch budget; and the requirements of outdoor mounting and commissioning. A cheap camera on poor cable with insufficient storage costs more in reality — because it does not produce the evidence it was bought to produce.

At ZEMAVO we supply and install CCTV systems alongside access control and fingerprint systems and server room and data centre fit-outs. Start with a site survey and a written job for each camera before choosing any model — get in touch to arrange the survey.