Centinel Safety
Real-time personal protective equipment (PPE) monitoring with computer vision.
The challenge
Checking that every worker wears a helmet, vest, gloves, goggles and boots — and uses the right tool for the task — does not scale with manual supervision.
The solution
Plant cameras feed detection models that track each person by zone, flag violations with false-alarm filtering and notify the supervisor's phone.
What I built
- Real-time capture and inference service with .NET, ONNX Runtime and YOLO models, with person tracking.
- Zone rules: which PPE each area requires, and debounced compliance events.
- React and Vite web app: live view, zones, events and reports.
- Push alerts to the mobile app and a read-only Android monitoring panel.
Technical details
- Detects distant people with 960 px models
- No false alerts from detection flicker
- Isolated database and deployment per product
- Linux installer