How Real-Time AI Computer Vision Improves Facility Security

Most UK facilities record video that nobody watches. A guard cannot attend to forty feeds, so incidents are found afterwards, if at all, and the footage becomes evidence for an insurance claim rather than a way to prevent the loss.

Computer vision changes what cameras are for. A model running on hardware at the site watches every feed continuously, detects the conditions that matter and raises an alert with a clip while there is still time to act. Done properly, it also produces the safety record that HSE conversations need.

Conditions, not identities

Object detection models such as the YOLO family identify people, vehicles and objects in each frame and track them across frames. That is enough to express most security and safety needs as rules: a person in a zone after hours, a vehicle in a loading bay too long, a forklift and a pedestrian in the same aisle, a worker without a hard hat in a live area, a fire door propped open.

None of these require knowing who the person is. Designing around conditions rather than identities keeps the system accurate and keeps the data protection analysis straightforward. Biometric identification of individuals is a different legal category in the UK with a much higher bar, and it is rarely needed for security or safety outcomes.

The legal frame in the UK

Video of identifiable people is personal data under UK GDPR, and the ICO expects a lawful basis, proportionality, signage, retention set to the purpose and a data protection impact assessment for systems that monitor people systematically. The Surveillance Camera Code of Practice applies to relevant public authorities and is treated as good practice by everyone else. The ICO’s guidance on AI adds expectations about explaining what an automated system does.

A deployment that starts with a written DPIA for the specific site and use cases, states plainly on signage what the system detects, masks public areas and neighbours where required and keeps footage on the premises tends to satisfy all three without difficulty.

Why processing stays on site

Streaming every camera to the cloud is expensive, slow and creates a large store of footage nobody wants to hold. Edge hardware such as NVIDIA’s Jetson modules runs detection at the site, so only events with short clips leave the premises, the system keeps working when the connection drops, and residency questions from customers or insurers are answered simply.

Existing cameras can usually be reused. A survey assesses each one’s resolution, angle and lighting against the intended detection, and only the gaps are filled.

False alarms decide whether anyone responds

Sites with motion-triggered alarms turn them off within weeks because foxes, rain, headlights and moving shadows fire them all night and keyholders stop responding. Object detection tuned to the site distinguishes a person at the fence from a fox and a vehicle in the yard from a tree in the wind.

The honest measure is the false alarm rate against the previous system, measured in a pilot on real events, with zones and thresholds tuned in the first weeks. An alert that carries a clip lets the responder confirm before anyone is sent.

A safety record for the HSE conversation

HSE guidance on workplace transport is clear that separating vehicles and pedestrians is the priority on sites where both move, and that near misses should be recorded. In practice they rarely are, so safety managers work from incident reports written after someone was hurt. A vision system logs near misses as they happen, by zone, shift and time, and gives the evidence for a layout change, a training intervention or a conversation with a contractor.

How a deployment runs

Survey and DPIA first. A pilot on a handful of cameras with false alarm rate measured. Rollout with alerts delivered to phones, the control room, the incident system or access control in the format each expects, followed by a tuning period on real events. A managed service keeps the hardware healthy, the models current and the reporting flowing to security and safety teams.