All work

Phone and seat‑belt violation detection

Computer vision that flags phone use and unbuckled belts from roadside cameras, in production across Saudi Arabia since 2018.

Client
Tahakom
Sector
Road safety / traffic enforcement
Country
Saudi Arabia
Period
2018–present
Status
In production, nationwide

How the system fits together

Violation detection system diagramInside the client network — no external connectivityOn-premise GPURoadside camerasIngestion and trackingDetection modelsphone · seat beltHuman reviewEnforcement workflowAudit trail: frame, model version, confidence

A computer-vision pipeline that analyses roadside camera frames, detects drivers using a handheld phone or driving without a seat belt, and hands evidence packages to the enforcement workflow with an auditable confidence trail.

The problem

Distracted driving and unbuckled belts are two of the most common contributors to road fatalities, and neither can be enforced at scale by officers alone. Tahakom needed detection that was accurate enough to stand up to review, and robust to glare, night conditions, tinted glass and vehicle variety.

What we built

We trained and hardened detection models on regional data, built the ingestion and tracking pipeline that turns raw frames into evidence packages, and delivered a review console so human operators confirm each case. Everything runs inside the client's own infrastructure.

How it is run

The system is monitored around the clock, models are re-evaluated against fresh samples on a fixed cadence, and every detection carries the frame, model version and confidence that produced it.

Details shared under NDA on request.

Nationwide
production deployment
24/7
continuous inference
Auditable
every decision traceable to a frame

Scope

  • Detection models for phone use and seat-belt state
  • Frame ingestion, tracking and evidence assembly
  • Human-review tooling and audit trail
  • On-premise deployment and monitoring

Technology

  • PyTorch
  • TensorRT
  • Python
  • Kubernetes
  • PostgreSQL
  • On-premise GPU

Planning a digital transformation, an AI pilot or a citizen-facing service?

Send us the brief or the tender documents. You will hear back from an engineer, not a salesperson, within two working days.