BlogAI
AI cameras: what they see,
what they miss, and what GDPR asks
An ordinary camera tells you what happened, afterwards. An AI camera tells you while it is happening. What works, what does not, and what must be sorted legally before you mount anything.
Recording versus understanding
A classic system writes everything to a hard drive and waits. Its value only appears after an incident, when someone scrubs through hours of footage to find thirty seconds. If nobody watches, the camera was a silent witness.
A system with recognition watches for you, constantly, and only calls when it sees something specific: a person in an area they have no business being in at that hour, a forklift parked on an escape route, a door left open, a vehicle entering the yard at night. The difference is not image quality, it is when you find out.
What it detects reliably
- A person present in a defined zone, within a time window
- Vehicles and number plates, at a yard or car park entrance
- Objects left behind or removed from a marked area
- Crowding, or more than a set number of people in a space
- Missing protective equipment, where it is required
- Crossing a virtual line, in one direction only
Where it gets things wrong, and why
No detection is perfect. Rain, fog, a spider web on the lens, a floodlight shining straight into it — all produce false alarms. A camera mounted too high sees the tops of heads rather than people, and confuses them.
Most problems come not from the model but from the installation: wrong angle, too little light at night, a detection zone drawn too wide. That is why tuning after installation is not optional, it is half the result. A week of adjustments cuts false alarms several times over.
A system that produces ten false alarms a day will be ignored within two weeks. Better to detect fewer things, correctly.
The legal side, without hedging
Video surveillance processes personal data, so it falls under GDPR. It is not forbidden, but it has conditions. There has to be a clear, legitimate purpose — securing property, keeping people safe — and the surveillance has to be proportionate to it.
In practice that means a handful of concrete things, and those are what get checked.
- Visible signage at the entrance saying who is monitoring and why
- Written notice to employees before the system goes live
- No cameras in changing rooms, toilets or rest areas
- A defined retention period for the footage, actually observed
- Access to footage limited to named people, with a log of who viewed what
- Cameras must not film public land or a neighbour property
We are not GDPR consultants. The list above is orientation; for your specific situation, check with a data protection specialist.
How to do it so it is not redone
On-site assessment
Decide what needs to be seen and why. The number of cameras and their positions follow from that, not the other way round.
Network first
Cameras sit on the network. Without proper cabling, a switch with power over cable and zone separation, the system will be unstable no matter how good the cameras are.
Install and tune
Angles, lighting, detection zones. Followed by a period of adjustment against real footage.
Paperwork and briefing
Signage, notices, who has access. Done at go-live, not when an inspection arrives.
What this is about, in full
Frequently asked questions
Is facial recognition legal in a company?
Biometric data has a special and much stricter regime than plain filming. For most companies, detecting people and behaviours solves the problem without going anywhere near it.
Can I use cameras to check up on employees?
Monitoring employee activity has its own conditions, stricter than securing property, and requires notice and proportionality. It is not solved by mounting a camera above a desk.
Will AI work with the cameras we already have?
Often yes. The analysis can run on a separate machine that takes the feed from your existing cameras. We check their resolution and positions first.
Where is the footage stored?
Locally, on a recorder at your premises, is the usual and easiest option to justify. Cloud is possible, but it adds further discussion about where the data ends up.