Video analytics: what security cameras can actually detect

• Line crossing: an alert when someone crosses a virtual line in one direction — over a fence, or through a gate after hours.
• Intrusion zone: an alert when a person or vehicle enters an area that should be empty, such as a storage yard at night.
• Loitering: someone stays in one place longer than you allow — at a door, by a till, in a car park.
• Left or removed object: something appears in or disappears from the scene and stays that way.
• Vehicles and plates: a vehicle detected in a zone, and its number plate read at a gate or loading ramp.
• People counting: how many go in and out of an entrance, for occupancy or evacuation.

What these have in common is that the system answers one specific question at one specific spot. It does not "understand" what is going on. Where each camera goes and what job it does is covered in the guide to choosing a CCTV system.

• Weather: rain, snow, fog and drops on the lens change the image; wind moves trees, flags, tarpaulins and cables.
• Light: headlights, sunrise and sunset, passing clouds and lengthening shadows. In the dark the image gets noisy and small shapes are lost.
• Scene: animals, insects and spiders in front of the lens, reflections in glass and water, a zone drawn too close to a road or pavement.
• Maintenance: a camera knocked out of position, a dirty lens, a failed light. The zone stays the same while the picture has changed.

The most dangerous false alarm is the repeated one: when the same zone fires every windy night, the operator learns to ignore it — including on the night it is real.

There is also the opposite failure: the missed alarm. Someone moving slowly, crawling, wearing dark clothes in the dark or using a blind spot may trigger nothing. That is why critical points combine image analysis with another form of detection, as the guide to perimeter protection explains.

Verification is short but specific: the operator opens the live image and the recording of the seconds before the alert, sees what caused it, and decides according to a written procedure.

1. What triggered it? A person, a vehicle, an animal, weather, light.
2. Are they authorised? Staff on shift, a scheduled delivery, a maintenance crew.
3. What are they doing? Passing through, searching, climbing, carrying something.
4. What action is set out? A warning, sending a guard, calling the police, informing the client.
5. Record it: what was seen, what was decided, when — so false alarms get fixed and incidents are documented.

Analytics is what makes remote guarding possible: nobody can watch dozens of screens all night, but someone can open the right camera the moment a zone fires. What remote monitoring can and cannot do, and when you need a guard on site, is covered in that guide.

In OVIT CCTV systems, visual verification is linked to the guard who acts, and on the Guardly platform the alert, the action and the report sit in the same incident record.

• What biometric data is: data from specific technical processing of a person’s physical characteristics that allows or confirms their unique identification, such as facial images (GDPR, Article 4(14)).
• The rule: processing biometric data for the purpose of uniquely identifying a person is prohibited (Article 9(1)).
• The exceptions: the prohibition lifts only in the cases listed in Article 9(2) — the first being the person’s explicit consent for a specified purpose.

A camera at an entrance, a till or a fence sees passers-by, customers and visitors who have given no consent at all. So for ordinary security needs, design the system without facial recognition and ask the supplier to confirm in writing that the function is switched off. A system telling "person" from "vehicle" is different from one telling you who the person is.

Cameras may not be used to monitor or evaluate employees (Hellenic Data Protection Authority, citing Article 27 of Law 4624/2019). That applies to analytics too: counting or loitering detection must not become a tool for checking on staff.

The GDPR defines personal data as any information about a person who can be identified directly or indirectly, in particular by reference to an identifier such as a number (Article 4(1)). A list of plates with entry times is therefore a list of personal data and needs a purpose, a retention limit and restricted access.

• Signs: inform people with a sign before the vehicle enters the camera’s range, naming the controller, the purpose and where rights can be exercised (Directive 1/2011, Article 12).
• Retention: footage is deleted within 15 working days at most if nothing happens; 30 days if an incident concerns the business, 3 months if it concerns a third party (Article 8). Apply the same rule to plate logs and alert snapshots.
• Counting: keep only the number you need, not images or people’s paths.

The Hellenic Data Protection Authority states that Directive 1/2011 still applies alongside the GDPR and Law 4624/2019. Permits, signs and employees are covered in the guide to CCTV and the GDPR for businesses.

1. Define the question: "alert me if a person enters the yard after 20:00", not "alert me to suspicious movement".
2. Draw the zone: away from trees, roads and pavements, with a person or vehicle filter and a minimum target size.
3. Set a schedule: the zone arms when the site closes, not while staff are working.
4. Test it: someone crosses the line by day and by night, in rain if possible, and you check the alert reaches the operator.
5. Review it: every month, which zones produced the most false alarms and why — then fix the zone, the lighting or the camera position.