PLATFORM
Canopy Platform Edge Devices AI Agents Privacy by design
INDUSTRIES
Industries overview Hotels & Hospitality Retail Smart Cities Construction Out-of-Home
PROOF
Case Studies Insights
TECHNOLOGY
Technology overview Edge Compute Data Sovereignty OAIC Compliance
COMPANY
About Us Contact
Sign in Get in touch
CASE STUDY

National builder eliminates PPE gate incidents

Edge-based PPE detection at a major Melbourne compound.

Case Study · Construction — National builder, Melbourne compound

A national builder was catching PPE non-compliance at the gate with manual checks. The checks were inconsistent, and the incidents were still happening.

Construction site gate with workers in hi-vis and hard hats

The problem

The builder's Melbourne compound had a gate warden who visually checked every person entering. The warden was busy, the shift change was crowded, and the data on non-compliance lived in a paper logbook. The WHS team could not see patterns; they could only see incidents after someone had already entered the site without the right PPE.

Our deployment

We installed a Canopy CU-2 unit at the gate, fed by a camera positioned to look at people as they approached the turnstile. The edge ran a PPE agent: hard hat, hi-vis vest, safety glasses. If a person approaching the gate was missing any of the three, the turnstile would not open. The WHS team got a count of near-misses by hour, not a list of names.

The result

0PPE gate incidents post-deployment
21 daysRecord offline operation
0Worker identification

The compound's network was unreliable — typical for a construction site on the edge of a city — but the edge unit kept working through outages. The WHS team now sees a daily report: how many near-misses, at which hour, at which gate. They do not see who.

Ruggedised edge unit mounted inside a site equipment cabinet
The CU-2 unit ran through network outages with a 21-day offline record.

Implementation details

The gate ran a Canopy CU-2 unit in a weatherised cabinet next to the turnstile. The camera was mounted at 3.2 metres, angled down at the approach lane. The PPE model was tuned on footage the builder provided during a two-week calibration period; we did not use any of that footage for general model training. The turnstile integration was via dry contact relay — a simple, fail-safe circuit that the site's electrician could wire in an afternoon.

When the edge detected a person without hard hat or hi-vis, it held the gate closed for 1.5 seconds, then opened if the condition cleared. If the condition persisted, the WHS officer received a Teams message with a zone and timestamp. No image was retained. The WHS team saw a count, not a photograph.

Before and after

The WHS team had logged around 12 near-misses per week on paper. The system counted around 25 per week, most of them at the gate. Incidents inside the compound fell to zero within a month.

The gate warden monitors a screen that flags exceptions instead of checking every person at a crowded shift change.

Ongoing support

The builder has since rolled the same pattern to three more compounds. We now provide a quarterly compliance report: near-miss counts by hour, by gate, by PPE type. The WHS team uses it to target inductions and signage. The edge discards frames, so there is no image to share.