PPE Compliance Monitoring System

The problem

PPE compliance is enforced by whoever happens to be watching. Safety officers cannot observe every area continuously, so lapses are found during audits or after an incident — both of which are too late to be prevention.

How it works

YOLO models process existing factory camera feeds and detect whether required protective equipment is being worn in each area, alerting safety officers in real time. OpenCV handles the video pipeline.

Inference runs on edge GPUs at the site rather than in the cloud. That was a requirement rather than an optimisation: the alert is only useful while the person is still in the area, and streaming continuous multi-camera video off-site to find out is both too slow and a considerable amount of footage to be moving around.

What shaped it

Alert fatigue was the main design risk. A system that fires constantly gets muted within a week, so detection was tuned to sustained genuine non-compliance rather than to every momentary frame where equipment was occluded.

Keeping video on-site also avoided turning a safety tool into a surveillance concern — the system reports compliance events, not continuous footage of individuals.

Outcome

Safety officers were alerted to non-compliance in real time rather than after the fact, and workplace safety incidents fell.