Use case
PPE Detection on Active Construction Sites: What It Catches
Hard hats and hi-vis are the easy part. The detections that matter are the ones a walk-around would never have caught, because they happen when the safety lead is on the other side of the site.
VuePoint
A contractor we work with in Arizona put it plainly: the toolbox talk happens at seven, and by ten the hard hat is on the toolbox instead of the head. Nobody is being reckless. It's hot, the hat is uncomfortable, the task takes four minutes, and the safety lead is two acres away dealing with something else.
That gap — between the policy and the middle of the shift — is the whole problem. It is not solved by more training, and it is definitely not solved by a monthly audit. It's solved by somebody noticing at the time.
What the system is actually looking for
PPE detection is often described as though it were one capability. It isn't — it's several detections with quite different reliability characteristics, and it's worth knowing which is which before you set expectations with a safety committee.
| Detection | How well it holds up | Why |
|---|---|---|
| Hard hat present or absent | Strong | High contrast, consistent shape, sits at the top of the silhouette where it's rarely occluded |
| Hi-vis vest | Strong | Colour and reflectivity make it distinctive in most lighting |
| Working at height without fall protection | Good, and the highest value | Harness detection is harder, but the context — person on a leading edge or roof — is very learnable |
| Eye protection | Moderate | Small, often occluded, easily confused with ordinary glasses at distance |
| Gloves, boots | Weak at realistic camera distances | Too small in frame to classify reliably. Treat claims about these with scepticism |
The detections a walk-around never catches
The value isn't in duplicating what a safety lead would see. It's in the hours and places they aren't.
- The four-minute task. Someone steps into an area to do something quick and doesn't gear up for it. Over by the time anyone walks past.
- Early morning and late afternoon. First hour and last hour, when supervision is thinnest and people are moving fastest.
- Deliveries and visiting trades. Subcontractors and drivers who weren't at your toolbox talk and are following their own employer's rules, or nobody's.
- The far corner. Every site has an area that's a long walk from the office and gets visited once a day.
- Work at height on a Friday afternoon. The combination that shows up in incident reports more often than anyone likes.
Same day, or it isn't worth doing
This is the part most implementations get wrong. A weekly PPE compliance report is a document. A same-day flag is a coaching conversation while the person still remembers the task.
The framing matters as much as the timing. Sites where this works treat it as a leading indicator — a count of exposures caught and corrected, trending down — rather than as evidence for disciplinary files. Sites where it's used punitively get one thing very reliably: people who have worked out where the cameras point.
How it connects to what you actually pay
PPE exposures are a leading indicator for recordables, recordables become claims, and claims drive your Experience Modification Rate — which is both an insurance multiplier and, on a lot of jobs, a prequalification threshold you have to clear to bid at all.
The chain is real but it is slow: EMR runs on a rolling three-year window that excludes the most recent year, so this year's improvement shows up in two to three years' rate. The faster effect is evidentiary — a disputed injury claim you can actually reconstruct. We've written up the mechanism and where to research it properly, including which sources to trust and the five questions worth asking your broker.
What is PPE detection?
PPE detection is video analytics that checks whether people in view are wearing required protective equipment — hard hat, high-visibility vest, and on some sites safety glasses, gloves or harness. The software identifies a person, then checks for the item, and logs the ones missing it. It is a compliance and safety-culture tool, not a disciplinary camera, and the difference matters a great deal to whether it survives contact with a workforce.
Evaluating PPE detection from the major platforms
PPE detection is now offered across most of the established camera and VMS platforms — Verkada, Avigilon, Hanwha and others — either natively or through an analytics integration. If you are already committed to one of those ecosystems, the sensible first move is to ask your existing vendor what they support, because running it on cameras you already own is cheaper than anything else you can do.
We are not going to publish a feature matrix comparing those products. We have not tested them, their capabilities change release to release, and a table of guesses would be both unhelpful and easy for the vendors to disprove. What we can give you is the set of questions that actually separates a working PPE deployment from one that gets switched off — and these are worth asking of us as readily as of anyone else.
Six questions to ask any PPE detection vendor
- 1
At what distance does it still work?
This is the question that decides everything and the one least often answered with a number. PPE detection needs enough pixels on a torso and a head. A camera covering a whole site from a mast will not do it. Ask for the maximum working distance for your camera and lens, not a general accuracy percentage.
- 2
Which items, and how reliably each?
Hard hats and hi-vis are high-contrast and detect well. Safety glasses, gloves and harnesses are small, often occluded, and materially harder. A vendor quoting one accuracy figure across all PPE types has not deployed it on a real site.
- 3
What happens with a hard hat carried, not worn?
A worker holding a helmet is the classic edge case, and it is common at breaks and during transitions. Ask what the system does with it. The honest answer involves a rate, not a denial.
- 4
Does it run on cameras we already own?
If yes, your first deployment costs no hardware. If it requires the vendor's cameras, you are buying an infrastructure programme, and that should be priced as one.
- 5
Does it need power and network at the position?
Most PPE-relevant areas — active work faces, material handling zones, access gates — are exactly where a construction site has no permanent infrastructure. Ask what happens when the answer is no.
- 6
Is it identifying people, or detecting PPE?
These are different products with different regulatory obligations. Several US states regulate biometric identification separately. Get the answer in writing, because it determines what you must tell your workforce and your works council.
What it doesn't do
- It doesn't make the site safe. It surfaces exposures. Somebody still has to act on them, and if nobody does, you've bought a log.
- It doesn't replace a safety programme. It's an input to one.
- It doesn't see through things. Occlusion, distance, weather and camera placement all bound what's detectable — a camera looking into low sun at 5pm is not going to classify anything usefully.
- It doesn't judge intent. It can tell you someone was on a leading edge without visible fall protection. It can't tell you whether they were clipped to something out of frame. Every detection is a prompt to look, not a verdict.
On sites without power or a network — which describes most of a job's early phase — this runs on solar and cellular units that move as the work moves, because the areas that need watching in month one are rarely the areas that need watching in month six.
Common questions
- What is PPE detection?
- Video analytics that checks whether people in view are wearing required protective equipment — typically hard hat and high-visibility vest, sometimes glasses, gloves or harness. It identifies a person, checks for the item, and logs those missing it. It detects PPE; it does not identify individuals, which is a separate technology with separate regulatory obligations.
- Does Verkada, Avigilon or Hanwha do PPE detection?
- PPE detection is offered across most established camera and VMS platforms, natively or via integration. If you already own one of those systems, ask your vendor first — running analytics on cameras you already have is the cheapest starting point. We deliberately do not publish a comparison matrix of their capabilities, because we have not tested them and those capabilities change release to release. The six questions above are what actually separate a working deployment from one that gets switched off.
- Does it work at night and in bad weather?
- Detection degrades in poor visibility for every camera-based system, ours included — that's physics, not a software limitation. Hard hat and hi-vis detection hold up well under IR at working distances. Anyone claiming unchanged accuracy in fog and heavy rain is describing a product that doesn't exist.
- Can it tell which subcontractor a person works for?
- Not by identity. Some sites approximate it by zone and time — if the concrete crew is in the east pour between 7 and 11, exposures there are attributable by context. That's an operational inference you draw, not something the system asserts.
- How many alerts should we expect?
- In the first fortnight, more than you want — that's normal and it's information. The count should fall as the site adjusts and as zones and thresholds are tuned to your actual layout. If it hasn't fallen after a month, the configuration is wrong or nobody is acting on the alerts.
- Do we have to tell workers?
- You should, and depending on where you operate you may be required to. Beyond the legal position, it works better: a safety system people know about changes behaviour, and one they discover by accident damages trust for longer than it improves compliance.
- Can this be used in a disciplinary process?
- That's your policy decision, not a technical one — but sites that use it that way tend to get avoidance rather than compliance. The programmes that work treat exposures caught as the metric, and drive that number down.
Want to see what your site would surface?
Tell us the layout, the trades on site, and where supervision is thinnest. We'll tell you what's realistically detectable from where the cameras can go.