Choosing security cameras for a workplace

The commonest disappointment with camera systems is not image quality on the monitor — it is discovering, at the moment the footage is needed, that it cannot answer the question being asked of it.

Specify per viewpoint, not per area

Before choosing hardware, decide what each position must achieve. The industry framework is useful because it maps directly to pixel density on the subject:

• Monitor — something is happening in this area. Low density is enough.
• Observe — how many people, moving which way, what are they carrying.
• Recognise — is this someone already known to you.
• Identify — footage usable by a third party who has never seen the person.

These requirements differ by an order of magnitude. A single wide camera covering a car park may monitor adequately and identify nobody. Entrances and till points usually need identify; general floor coverage usually does not.

Resolution is not the whole answer

An 8 MP camera covering a wide area can deliver fewer pixels on a face than a 2 MP camera correctly positioned on a doorway. What matters is pixels per metre at the subject, which is a function of resolution, lens and distance together.

Higher resolution also costs storage and bandwidth, and — importantly — performs worse in low light, because smaller sensor pixels gather less light. More megapixels is not a free upgrade.

Low light is where systems disappoint

Most incidents worth recording happen when the lights are off. Two things matter more than the headline resolution.

Infrared range is quoted optimistically and measured in ideal conditions. Treat published figures as a ceiling, not a specification.

Backlight ruins more footage than darkness does. A camera pointed at a glass entrance during the day produces a silhouette and nothing else. Wide dynamic range handling matters at these positions, and so does mounting angle.

Storage: work backwards from retention

Recording capacity is a calculation, not a product feature. It depends on camera count, resolution, frame rate, codec efficiency and — the variable people forget — how much motion there is, since modern codecs use far less space on a static scene.

Decide the retention period first, calculate from there, and add margin. Adding cameras later without adding disks silently shortens retention across the whole system, which is how businesses discover their thirty days quietly became eleven.

KVKK: retention and notification

Camera footage is personal data and the obligations are specific.

• Visible notification at the monitored area, before entry.
• A defined retention period with a reason you could state if asked. Thirty days is common for general premises monitoring. Indefinite retention because nobody configured it is a finding, not a policy.
• Restricted access to recordings, with a record of who viewed what.
• Proportionate placement. Cameras in areas where staff have a reasonable expectation of privacy — rest areas, changing rooms — are not defensible.

These are design decisions, not paperwork to be added afterwards.

Do not put cameras on the office network

Camera firmware is, as a category, not well maintained. Devices frequently ship with weak default credentials and receive updates rarely.

Cameras belong on their own VLAN with no route to the office network and no direct internet access. Remote viewing should be published through a controlled path — not by forwarding a port to the recorder, which is how a great many recorders end up indexed on public device search engines.

Before you buy

Walk the site and list the viewpoints. For each, write down what the footage must prove, the lighting at the worst time of day, and the distance. Then specify.

That list is a better basis for a quotation than a camera count, and it is what distinguishes a system that answers questions from one that merely records.