Case Studies  /  Civil Construction
CASE STUDY · CIVIL CONSTRUCTION

54% fewer acknowledged findings across a civil construction programme.

A group of civil construction contractors put Invigilo on the cameras they had already installed. The work includes deep shafts and tunnel headings. In the second year, acknowledged safety findings came in 54% below the first.

CIVIL CONSTRUCTION CONTRACTORS
Results at a Glance
54%
Fewer acknowledged findings, year two against year one
CIVIL CONSTRUCTION CONTRACTORS
2mo
From first camera access to full programme coverage
CIVIL CONSTRUCTION CONTRACTORS

Most of the work happens where nobody is watching

Deep excavation, access shafts, tunnel headings. Tens of metres down, off any patrol route, and usually the part of a civil programme with the least supervision per worker.

Most safety systems assume someone can see the work. Below ground that assumption breaks, and the hazards themselves change shape. A supervisor cannot walk a heading, a shaft and a station box inside one shift.

These contractors were running several worksites at once, with different layouts and different subcontractors on each. Five risks came up repeatedly.

A
Falls into open shafts
The drop is on the inside. Detection trained on rooftop and facade footage read these scenes badly until it was retuned on site.
B
No room to step aside
In a bore the gap between plant and wall can be a few centimetres. A worker caught in that space has nowhere to go.
C
Exclusion zones that move
Cutterhead work, segment handling and grouting each close an area for a while, then release it. Fixed signage cannot keep up.
D
Barricades that do not go back
They come out for access. Sometimes they stay out for a shift, and nobody logs it.
E
Lone workers out of signal
Wearables and phone check-ins depend on coverage. Underground there often is none.

The cameras were already on site

Every worksite already ran CCTV, mostly for security and progress photos. Nobody added a camera or pulled new cable. Invigilo took the existing RTSP streams and processed them on site, at the edge. Underground the uplink drops often enough that anything relying on it would be useless.

Full coverage took two months. A single site is usually live inside 24 hours. The difference here was volume: cameras across several concurrent worksites, each needing its zones drawn and its rules set against footage from that specific environment. Underground scenes in particular needed a few weeks of real footage before the thresholds were any good.

Edge proximity and open penetrations Plant and worker separation Exclusion zone entry Missing barricades Lone worker
SURFACE COMPOUND SHAFT HEADING TENS OF METRES DOWN EXISTING CAMERA, ALREADY IN PLACE DETECTION RUNS ON SITE, NO UPLINK NEEDED ILLUSTRATIVE SECTION, NOT SITE DATA
Illustrative section: a compound at surface, the shaft and heading below it, and the cameras that were already covering each

Existing CCTV

Cameras were already in place across compounds, shafts and headings. No new hardware was needed for standard coverage.

Edge processing

Detection runs on site. Alerts keep working through the connectivity gaps that are normal below ground.

Rules per zone and phase

An area can be dangerous during a segment lift and ordinary an hour later. Rules are set per zone and switched with the phase of work.

Alerts routed by severity

They go to the supervisor responsible for that area, on WhatsApp or Teams, with the clip attached.

What changed across the programme

The number took a year to move. The change in how supervisors worked showed up much earlier.

54% fewer acknowledged findings

Second year of the programme against the first.

Faster response to what was found

An alert goes to the supervisor for that area while the work is still happening, instead of waiting for the next walkaround.

Supervisors changed how they work

They act on alerts themselves. Clips get used in briefings. Where something keeps happening, they take it to the subcontractor responsible instead of raising it with everyone.

What made it stick

Detection on its own changes very little. What mattered was who got the alert and what they did with it.

It ran on every shift

Several worksites cannot be covered by walking them. Nights and weekends get the same attention as a Tuesday morning.

It reached someone who could act

Alerts went to the supervisor for that area, with the authority to stop the work.

It kept the evidence

Every finding has its clip. That is what makes a toolbox talk land and lets a repeat issue be traced to a crew.

It showed patterns across sites

When the same thing goes wrong on three headings, that usually points at how the work is set up.

See it on your own worksites

Start with the cameras already covering one shaft or heading. A single site is usually live within 24 hours of access.

Get a Demo Talk to Sales