Case Study — AI CCTV for Construction | SiteVision AI
Case Study
CS-2024-07

AI-Powered CCTV  ·  Construction  ·  NX Witness

Seeing the Site
the Way a Safety
Officer Can't

How we deployed an intelligent CCTV and AI analytics system across an active high-rise construction site — reducing safety violations by 74% in 90 days.

Client

Mid-size Construction Firm

Project scale

28-storey development

Duration

14 weeks

Platform

NX Witness + AI Manager

01

The Challenge

The client was a growing construction firm managing a 28-storey mixed-use development in an urban centre. With over 120 workers on site at peak, spread across multiple floors, a tower crane operation, and a busy site entrance shared with public footpaths — their existing CCTV system was purely reactive. Cameras recorded. Nobody watched. Incidents were reviewed after the fact, not prevented.

Following two near-miss incidents in a single quarter, the project director needed a system that could act as a permanent, tireless safety presence across the entire site.

PPE Non-compliance

Workers removing hardhats and hi-vis vests in active zones, undetected until incident review.

Exclusion Zone Breaches

Crane lift areas and excavation perimeters regularly accessed without authorisation.

After-hours Intrusion

Perimeter hoarding breaches overnight leading to tool theft and site liability concerns.

No Visibility on Progress

Management had no reliable way to remotely track site progress or validate subcontractor activity.


02

Our Solution

We designed a layered CCTV and AI analytics deployment using NX Witness as the VMS backbone, with Nx AI Manager handling model orchestration. The system was built in four integrated phases.

Phase 1 — Infrastructure & Fixed Camera Network

Installed fixed IP cameras at all perimeter hoarding points, the site office block, welfare facilities, and the tower crane — covering the counter-jib, operator cabin approach, and crane base exclusion zone. Cameras selected for wide dynamic range to handle Singapore's harsh contrast between shade and direct sun.

Perimeter  ·  Office  ·  Tower Crane

Phase 2 — Mobile CCTV Deployment

Deployed two rapid-deploy mobile CCTV units on battery and solar power for internal site coverage at the active working floor and hoisting bay. Units were repositioned weekly as construction progressed upward — giving us eyes on the most dangerous active work zones at all times.

Mobile Units  ·  Hoisting Bay  ·  Active Floor

Phase 3 — NX AI Manager Integration & Model Tuning

Deployed and configured three AI analytics pipelines on NX Witness: PPE/hardhat detection, person & vehicle detection, and perimeter intrusion. Critically, we spent two weeks tuning — drawing precise ROI polygons per camera, calibrating confidence thresholds to the site's specific PPE colour palette, and establishing day/night rule scheduling to eliminate false alarm fatigue.

AI Analytics  ·  ROI Tuning  ·  Rule Scheduling

Phase 4 — Exclusion Zone Enforcement Layer

Built a dedicated exclusion zone rule set on top of existing person detection — flagging any person detected within the crane lift radius during active lifts, within 2m of excavation edges, and within scaffold base zones without simultaneous hardhat detection confirmation.

Exclusion Zones  ·  Compound Rules  ·  Crane Safety

03

AI Modules Deployed on NX Witness

Analytics Module Coverage Key Configuration
PPE Detection 8 cameras — all active work zones Confidence set to 68%; yellow & orange hi-vis classes separated; 75-sec cooldown per zone
Perimeter Intrusion 6 cameras — full hoarding perimeter Direction-of-travel filter (inbound only); night mode high sensitivity 19:00–06:00; linked to mobile PTZ
Person & Vehicle Detection 4 cameras — entrance, hoisting bay, crane base Separate event streams per class; vehicle proximity alert within 5m of active crane zone
Exclusion Zone Rules Crane lift area, excavation, scaffold base Compound rule: person detected + no hardhat confirmed = immediate alert; crane lift schedule-linked

04

Results at 90 Days

74%

Reduction in recorded PPE non-compliance incidents

0

After-hours perimeter breaches in the final 6 weeks of review period

91%

Reduction in false positive alerts after ROI tuning vs. week one

Beyond the safety metrics, the client's project director gained remote visibility into site activity through the NX Witness mobile app — able to pull live feeds and AI event logs during off-site management meetings. The system also provided timestamped site progression footage used to resolve a subcontractor attendance dispute, saving the firm a potential contractual penalty.

"Before this system, we were reviewing cameras after something went wrong. Now the system tells us when something is about to go wrong. That's a completely different posture for site safety — and honestly, it's changed how our safety officer does his job."

Project Director  ·  High-rise Residential Development


05

What We Learned

  • Out-of-the-box AI confidence thresholds are tuned for generic environments. On a construction site, two weeks of dedicated tuning delivers more value than the initial hardware deployment.
  • Mobile CCTV repositioning must be planned into the project schedule — not treated as an afterthought. Weekly repositioning briefings with the site manager kept us ahead of the active work face.
  • Alert fatigue is the silent killer of AI CCTV projects. A system generating 200 false alarms a day trains operators to ignore everything. Ruthless ROI definition and cooldown tuning is not optional.
  • Compound rules — stacking two analytics conditions simultaneously — dramatically improve precision over single-model triggers. Our crane exclusion zone rule only fires when person detection and hardhat absence are both confirmed at the same time.
  • The client relationship deepens when you build a weekly AI performance review into the contract. It makes the data visible, builds trust, and surfaces tuning needs before they become problems.
Planning a construction CCTV deployment? We bring NX Witness expertise, AI analytics experience, and a structured methodology built specifically for active construction environments.
Get in touch →
Scroll to Top