PPE Detection Cameras on Jobsites: A First Look at Benefits and Limits

PPE detection cameras on jobsites can add another alert source to a busy construction site. As of August 2026, UniFi Protect's current feature can flag people who appear to be missing a helmet or safety vest. It is not a general PPE-inspection engine, and it does not evaluate eye protection, gloves, hearing protection, respirators, footwear, harnesses, fit, or the requirements of a particular task.

It is not an Occupational Safety and Health Administration compliance system. A camera cannot perform the hazard assessment, select properly fitting equipment, train the crew, supervise the task, or replace the competent people responsible for the safety program.

What the current Protect release actually detects

Ubiquiti's Protect 7.2.105 release notes identify the supported PPE classes as helmets and safety vests. The current release requires Protect 7.2, compatible camera firmware 5.4, and AI Key firmware 2.2 or newer. Check those requirements again before deployment because this is a new, fast-changing feature.

That can be useful at controlled entry points, material yards, shop floors, and other locations where a camera has a consistent view of helmets and safety vests. It can also help a supervisor find footage worth reviewing instead of watching every camera continuously.

The feature depends on the wider Protect design. Camera angle, lighting, distance, clothing, congestion, and a worker or object blocking the camera's view all affect what the system can see.

The model is an alert source, not a safety decision-maker

Ubiquiti's current AI Key FAQ warns that AI results can occasionally be inaccurate. Only events recognized and processed by the system appear in advanced results. A processing queue can delay output, and excess detections can be discarded when capacity is exceeded.

False negatives matter because a missed alert can create false confidence. False positives matter because a safety manager who receives too many useless notifications will stop trusting the system. Pilot the feature in one controlled area, review real footage, and tune the response before expanding it.

Ubiquiti does not publish jobsite sensitivity or false-alert rates in the cited product pages, and GTZ will not invent them. Until a site-specific pilot produces measured results, treat PPE detection as an available alert feature rather than a validated safety control. Ubiquiti also says AI Key processes a maximum of 1,000 detections per hour, queues up to 200 events, can delay results by as much as 15 minutes when the queue is full, and discards excess detections. Treat the word "real-time" as a vendor claim that must be tested under actual site traffic.

Do not automatically discipline an employee from one AI alert. Review the video, confirm the policy and context, and follow the company's established safety and personnel process.

OSHA's obligations still belong to the employer

OSHA's construction PPE rule requires employers to ensure protective equipment is suitable, properly fits affected employees, and is provided, used, and maintained where hazards require it. The rule does not transfer those duties to a camera vendor.

The system may support observation and documentation. It does not identify every hazard, determine which PPE the task requires, or know whether a specific exception is allowed. A properly fitting hard hat also cannot be evaluated reliably from every camera view.

A camera can flag what it sees. The employer still owns the hazard assessment, training, supervision, and response.

Where the technology can help

A useful deployment begins with a narrow operational question, not "watch the whole jobsite for safety." Better examples include:

• Alert when a person enters a clearly marked area without the configured helmet or safety-vest condition
• Review recurring patterns at a shop or warehouse entrance
• Give a safety manager a faster pointer to footage worth reviewing
• Document whether an alert was acknowledged and investigated
• Compare alert quality across angles before buying more hardware

The response matters more than the notification. Decide who receives the alert, how quickly it must be reviewed, what counts as confirmation, and what happens when the AI is wrong.

Use a pilot scorecard before buying more hardware

Define the test on paper before turning on notifications:

• The exact zone and whether the test covers helmets, safety vests, or both
• Camera height, angle, lighting, distance, and expected worker traffic
• How many known compliant and known noncompliant examples will be sampled
• Observed false positives, false negatives, notification delay, and discarded events
• The person who reviews each alert and the required response time
• Worker notice, footage retention, personnel-use rules, and stop criteria

The pilot should end with a written go, adjust, or stop decision. A few successful demonstrations from a quiet doorway do not establish performance across a dusty, backlit, congested jobsite.

Hardware and camera compatibility still matter

AI Key works from smart detections supplied by compatible Protect cameras and related accessories. Ubiquiti says older G3 or third-party ONVIF cameras require AI Port before AI Key can process their streams. An AI Key alone is not a universal upgrade for every legacy camera.

Even a compatible camera may be wrong for the position. Wide overview views are useful for awareness but can leave too few pixels on a worker to classify equipment reliably. Backlighting, rain, dust, trailers, lifts, and other workers can hide the item the system is trying to see.

Start with an evidence-quality image at the actual detection area. Then evaluate the AI.

Privacy and worker trust need a written answer

A construction company should tell workers what the camera system is used for, who reviews alerts, how long records are kept, and whether footage can be used in personnel decisions. Bargaining agreements, employment rules, insurance requirements, and project-owner policies may affect the program. Confirm those questions with the appropriate legal, human-resources, safety, or labor advisor.

Do not quietly turn a theft-prevention camera into an employee-surveillance program without governance. That is how a potentially useful safety signal becomes a trust problem.

Use PPE detection cameras on jobsites as another control, not the control

PPE detection is most useful inside a layered program: clear site rules, properly selected equipment, training, competent supervision, physical signage, incident reporting, and camera alerts that support human review.

GTZ sells and installs UniFi Protect and the AI hardware discussed here, so we have a commercial interest in the platform. We can design and test the camera and network layer through our physical security service. We do not certify OSHA compliance or replace the client's safety professionals.

For a contractor moving crews between Pueblo and Colorado Springs, the sensible first step is a limited pilot tied to one defined hazard-control process. The evidence-verification article explains the adjacent record-handling tools, and our construction work keeps the technology inside the way the job actually runs.