Real-time AI vision system
Using AI to streamline FDA Title 21 compliance and reduce review cycles
A leading offshore oil platform operator set out to reinforce safety compliance across its remote sites by adopting AI-based monitoring systems. In particular, they wanted to ensure that all personnel consistently wore hard hats in designated zones — a critical component of both worker safety and regulatory adherence.
Their goal was to move from reactive safety enforcement to a proactive, real-time approach using automated visual detection.
Ensuring proper protective gear usage on offshore platforms presented several challenges:
Manual Gaps: Supervisors couldn’t monitor all activity across all areas, leaving room for overlooked violations.
Delayed Response: Video footage was only used after incidents occurred, limiting the ability to prevent risks in the moment.
Tough Environments: Bandwidth limitations and harsh environmental conditions made cloud-only solutions impractical, requiring something lightweight and reliable on-site.
To address these issues, we developed and deployed a real-time AI vision system focused specifically on hard hat detection. Here’s how the solution came together:
Hard Hat Recognition: Using YOLOv8 and PyTorch, we trained models to detect whether individuals in camera footage were wearing hard hats, triggering alerts when non-compliance was detected.
Real-Time Alerts: The system delivered instant notifications to supervisors whenever a potential violation occurred, enabling immediate corrective action.
Accurate in Challenging Settings: The model was fine-tuned to handle varied lighting, movement, and obstructions common to offshore work environments.
Edge-Ready Design: For platforms with limited connectivity, the solution was deployed on edge devices — allowing all processing to happen locally without the need for continuous cloud access.
Adaptable Framework: The detection system was built to be extensible, enabling the client to expand into other forms of visual compliance in future phases without starting from scratch.
Simple Oversight Interface: A lightweight monitoring dashboard provided safety personnel with a real-time overview and historical records of flagged events.
With AI assisting the screening pipeline, clinics participating in the national program saw a significant reduction in the time it took to triage images.
The solution is now a key part of how diabetic eye screenings are scaled across regions with limited access to specialists.
By automating hard hat compliance checks, the client transformed their approach to safety — moving from passive monitoring to proactive enforcement. The system enabled them to catch violations early, respond quickly, and maintain a consistent safety culture without increasing headcount.
This project demonstrates how AI can be used not just to observe, but to actively safeguard teams in even the most remote and high-risk environments.
