PPE compliance with AI video: PPE detection in context

What PPE compliance is

Personal protective equipment (PPE) is the last line of protection between a person and a hazard that could not be removed or engineered out: a hard hat against falling objects, goggles against sparks and splashes, gloves against cuts and heat, a high-visibility vest where vehicles move, a harness where a fall is possible. PPE compliance is the question of whether that equipment is actually worn, the right equipment for the place and the job, and worn the way it protects. It breaks down in four recurring ways.

Equipment missing for the zone

Each zone of a site has its own requirements, set in the site's risk assessment: hard hats under overhead cranes, hi-vis vests on the dispatch yard, goggles at the grinding line. A person who walks from an office corridor into one of these zones has to put the equipment on at the boundary. The step is easy to skip when the visit is meant to be short.

Equipment wrong for the task

The same spot on the floor can call for different equipment depending on what a person is doing there. Walking past a welding bay is not the same as welding in it; carrying a box past a press is not the same as clearing a jam inside it. A rule written only for the zone misses this: it either demands too much from everyone, or too little from the person doing the hazardous task.

Carried, not worn — or worn the wrong way

A hard hat clipped to a belt, goggles pushed up on the forehead, a vest over the arm, a harness worn but not attached to an anchor: the equipment is present, so a quick glance says the person is covered, but it protects no one. For harnesses this matters most at height — see our guide on work at height and restricted zones.

Patterns behind the gaps

PPE gaps are rarely spread evenly. They gather at particular places and times: the end of a hot shift, a zone boundary with no equipment station, a task that is quicker without gloves, a dock where visiting drivers step out of their cabs. One missing hard hat says little; the same gap at the same place and time, week after week, points at a cause in the process or the layout — and that is what can be fixed.

Why traditional methods miss PPE gaps

Many sites have PPE rules, signs at the entrance, training and a PPE checklist. Gaps still slip through, because the checks are periodic and the gaps are momentary.

• Spot checks and audits see a moment. A safety walk observes one area for a short time, and people put their equipment on when they see the safety officer coming. The goggles pushed up at the grinding line on the night shift are not on the audit route.
• Checklists record that equipment was issued, not that it is worn. A PPE checklist or form confirms that hard hats are in stock and vests were handed out. It cannot show what happens on the floor an hour later.
• Manual CCTV review doesn't scale. The cameras are usually recording. But nobody can watch every zone on every shift, so footage is reviewed after an incident — not to find the gaps that came before it.
• Rule-based PPE detectors fire without context. A detector that only answers "is there a hard hat on this head?" flags the visitor in a corridor where no hard hat is required, the worker on a break outside the zone, and the person who has just stepped across the boundary to fetch one. It does not know the zone or the task. The result is a stream of alarms for normal situations — and a team that learns to ignore them.

What these methods lack is the judgment a safety professional applies on sight: where is this person, what are they doing, and what does that place and that task require?

How vision-language models make it solvable

Primarch builds on and enhances vision-language models (VLMs): models that read video and language together. That changes what a PPE check can be. Instead of ticking a box for each item on each head, the model reads the scene the way an experienced safety professional would — and does it on every camera, on every shift.

• It judges missing equipment against the zone and the task. The question is not only whether a hard hat is visible, but whether this place and this activity call for one. A visitor walking a marked route and a welder at work are not the same thing.
• It reads the action, not just the item. A VLM reads a window of time rather than a single frame, so it can tell a person welding from one walking past the bay. At height, it catches harness-free work at the level of the action itself.
• It uses context instead of fixed rules. Reading the scene in context is designed to remove the biggest source of false alarms in rule-based PPE detection: rules firing without context.
• It writes the finding in plain language, with a clip. The result is a sentence a supervisor can read — what was missing, where, during which task — with a time-stamped evidence clip.
• It chains events by shift, zone and time. Single events are recorded and alerted for review; chained together, they show where and when the same gap repeats, and help diagnose why: an attention lapse, or a process gap.

The useful question is not "is there a hard hat in this frame?" It is "does this person, in this zone, doing this task, have what the task requires — and are they wearing it?" The answer needs the zone, the action and the equipment together.

Coverage depends on the cameras. A zone can be checked only where a camera sees it, and whether a small item such as safety goggles can be read depends on how close the camera is and how sharp the image is. Which equipment is required in which zone comes from the site's own risk assessment and is set up together at onboarding.

What to look for in a solution

Whichever vendor you talk to, these are fair questions to ask about a system meant to check PPE compliance with video:

• Works on your existing cameras. Can it connect to the IP cameras you already have, or does every zone need new hardware?
• Understands zone and task. Can your requirements per zone and per task be set up, and does it tell a person passing through from a person doing the hazardous work?
• Tells worn from carried. Does it notice a hard hat on a belt, goggles on the forehead or a harness that is not attached?
• Handles false alarms openly. Ask how it avoids flagging people where no equipment is required, how findings are reviewed by a person, and how single events are separated from repeating patterns.
• Explains repeats. Can events be grouped by shift, zone and time, and does it help separate an attention lapse from a process gap?
• Produces reports your safety team can use. Findings written in regulatory language, corrective actions with owners, and zone-level summaries for the safety committee and audits.
• Respects privacy. Is footage anonymized? Is there an on-premise option where footage never leaves the site? Does it meet GDPR and other data-protection requirements, and does it report situations rather than individuals?
• Can be piloted on your own floor. One zone, a few of your cameras, your own equipment rules — not a demo video.

Example use cases

The walk-throughs below are illustrative examples of how a finding is built. They are not customer cases and contain no performance figures.

Example 1: a visitor and a welder in the same hall

1. 01

   Moment

   A fabrication hall, 10:40. A camera covers the welding bays and the marked visitor walkway beside them. Whether face protection is lowered can be read only where the camera shows the person's face, and is confirmed per site.

2. 02

   What the camera shows

   A visitor walks along the marked walkway in a hi-vis vest, without a welding mask. In bay 3, a person is welding with the mask lifted, sparks visible in front of their face.

3. 03

   The finding + evidence clip

   "Bay 3: welding in progress with the face protection raised. The person on the visitor walkway is outside the bay and was not flagged." Clip attached.

4. 04

   Who is alerted

   The shift supervisor receives an instant notification; the event goes into the safety team's weekly digest.

5. 05

   What the shift and zone pattern shows

   Similar events in bay 3 cluster at the start of short tack-welding jobs, when the mask is lifted to line up the part.

6. 06

   Corrective action

   A recommendation for the production manager: review the tack-welding step so parts can be lined up without lifting the mask, for example with an auto-darkening option or a fixture.

Example 2: grinding without goggles

1. 01

   Moment

   A grinding line in a metal plant, 22:15, late in the evening shift.

2. 02

   What the camera shows

   An operator grinds a part without goggles. The goggles lie on the bench beside the grinder.

3. 03

   The finding + evidence clip

   "Grinding line 2: grinding without goggles; the goggles are on the bench." Clip attached.

4. 04

   Who is alerted

   The evening shift supervisor, instantly.

5. 05

   What the shift and zone pattern shows

   Asked in plain language — "show goggle findings at grinding line 2 this month" — the event memory returns a short list, nearly all in the last hours of the evening shift.

6. 06

   Corrective action

   Diagnosis: a process gap rather than one careless person — operators report that the goggles fog up as the hall warms. Recommendation: anti-fog goggles and a ventilation check, assigned to the maintenance manager.

Example 3: a driver steps onto the dispatch yard without a vest

1. 01

   Moment

   The dispatch yard of a distribution center, 15:20. The yard is a hi-vis zone; a camera covers the dock apron.

2. 02

   What the camera shows

   A visiting truck driver climbs down from the cab and walks across the apron toward the dock office without a vest, while forklifts load the next trailer.

3. 03

   The finding + evidence clip

   "Dispatch yard: a person on foot crossed the vehicle area without a high-visibility vest while forklifts were working." Clip attached.

4. 04

   Who is alerted

   The dock supervisor, instantly; the safety team in the weekly digest.

5. 05

   What the shift and zone pattern shows

   Flagged events on the yard involve people walking from parked trucks to the office, during the afternoon dispatch window.

6. 06

   Corrective action

   A recommendation for the dispatch manager: hand out vests at the gate and mark a walkway from the truck parking to the office. For the wider vehicle–pedestrian picture, see our forklift and pedestrian safety guide.

Example 4: a harness worn but not attached

1. 01

   Moment

   A mezzanine edge in a warehouse, 09:05, during a lighting repair. A camera sees the mezzanine edge and the anchor rail. Whether a harness is attached can be read only where the camera shows the attachment, and is confirmed per site.

2. 02

   What the camera shows

   A technician works at the open edge wearing a harness; the lanyard hangs loose and is not clipped to the anchor rail.

3. 03

   The finding + evidence clip

   "Mezzanine edge: work at height with the harness worn but not attached to the anchor." Clip attached.

4. 04

   Who is alerted

   The maintenance supervisor, instantly.

5. 05

   What the shift and zone pattern shows

   No similar events in the last 30 days, so it stays a single event in the record rather than a pattern. The event also counts as a near miss — see our guide on near miss detection.

6. 06

   Corrective action

   A recommendation with the regulatory reference: confirm attachment before work at height starts, as part of the job's permit or pre-task check.

How Primarch's OHS Expert helps

This section is about our product. The OHS Expert is the Primarch expert module that reads the floor like a safety professional, on the cameras a site already has. For PPE compliance it:

• Checks PPE in context. It judges missing equipment against the zone and the task: a visitor and a welder are not the same thing.
• Checks hard hats, goggles, gloves and vests per line and task, and hi-vis and other equipment zone by zone, for example across a dispatch area.
• Catches harness-free work at height at the level of the action itself.
• Narrates each finding in natural language, with a time-stamped evidence clip, and recalls past repeats of the same scene from memory.
• Chains events by shift, zone and time and diagnoses why they repeat — an attention lapse or a process gap.
• Writes corrective actions in the language an inspector would use, assigns owners, and prepares regulation-referenced records, weekly zone safety scores and executive summaries.
• Needs no rule set. It comes pre-configured with OHS domain knowledge and regulation; site-specific thresholds are tuned together at onboarding, and situations nobody wrote a rule for are still caught through scene understanding.

Because it reads the scene in context, it is designed to remove the biggest source of noise in classic systems — rules firing without context — and multi-stage validation and temporal noise suppression are added on top.

Putting it to work

1. 01

   Connect

   Connect your existing RTSP cameras; no new cameras or special hardware.

2. 02

   Pilot on your own floor

   Pilots typically start with a single zone and a few cameras — a welding area or a dispatch yard, say. Your equipment requirements and thresholds are set up together, and you review the findings with your safety team.

3. 03

   Roll out

   Scale facility-wide as the system proves out, with dozens of streams watched in parallel.

• Alerts, digests and dashboards. Instant notifications, periodic digests and role-based dashboards, so each finding reaches the person who acts on it.
• Ask the past in plain language. Findings are written into a queryable event memory: "Which PPE findings were recorded on the dispatch yard last week?"
• Records you can defend in an audit. Every finding carries its time-stamped clip; reports carry the regulatory reference and the corrective action with its owner.
• Privacy by design. Footage is anonymized in a GDPR-compliant way, and the system is designed to report risky situations and process gaps, not individuals. With on-premise deployment, footage never leaves the facility.