The brief
Sentinel is a computer-vision-based PPE monitoring system designed to automatically identify whether workers are wearing required personal protective equipment.
The system analyses images and camera feeds to detect workers and safety equipment such as helmets and safety vests. When required PPE is missing, the system can identify the violation and surface it through the application’s alerting workflow.
The goal is to turn workplace safety compliance from a manual inspection task into a continuously monitored computer-vision workflow.
Automated safety compliance
Workplace safety often depends on workers consistently wearing the required protective equipment. Manually monitoring large sites, multiple workers, and multiple cameras is difficult and does not scale well.
Sentinel provides an automated visual monitoring layer.
Camera feeds can be analysed continuously, allowing the system to identify workers and determine whether expected PPE is present. Instead of requiring a person to watch every camera continuously, the system can highlight situations that require attention.
Real-time PPE detection
At the core of Sentinel is a trained object-detection model capable of identifying PPE-related objects from images and video.
The system can detect workers, safety helmets, and safety vests, and use those detections to identify missing PPE.
From detection to violation
Detecting an object is only part of the problem. For workplace safety, the important question is whether a worker is compliant.
Sentinel uses detected PPE information to identify violations such as a worker without a helmet or without the required safety vest. These violations can then be surfaced through the application’s alerting workflow.
Multiple camera sources
The system supports multiple camera feeds, providing a foundation for monitoring different areas of a workplace simultaneously.
Sentinel provides a common computer-vision layer capable of processing available feeds and surfacing PPE compliance information centrally.
Trainable detection pipeline
Sentinel includes dataset configuration, model training, testing, and inference workflows, allowing the detection model to be trained and evaluated against PPE-specific data.
GPU acceleration is supported where compatible hardware is available, helping reduce training time and improve inference performance.
More than a model
Sentinel connects the complete workflow:
Camera / Image Input → Detection → PPE Compliance Analysis → Violation Detection → Alerts → User Interface
This makes the system a complete safety-monitoring prototype rather than simply a machine-learning experiment.
Web and desktop interfaces
The web interface supports image, video, and live-camera workflows, while the desktop application provides a local interface for running the detection system directly.
This flexibility makes Sentinel suitable for development, demonstrations, and different deployment environments.
Practical computer vision
Real-world safety monitoring introduces changing lighting, camera angles, multiple workers, partial occlusion, background clutter, and varying distances from the camera.
Sentinel brings model inference together with camera handling, compliance logic, alerting, and application interfaces to create a practical end-to-end monitoring workflow.
The idea behind Sentinel
Safety teams should not have to rely entirely on someone noticing a violation at exactly the right moment.
Sentinel adds an automated visual monitoring layer that can continuously observe PPE compliance and surface potential violations.
See the risk. Detect the violation. Improve workplace safety.
Project highlights
- Real-time PPE object detection
- Helmet and safety-vest compliance monitoring
- Multi-camera support
- Image, video, and live-camera workflows
- Violation detection and alerting
- Custom model training and testing pipeline
- GPU/CUDA support
- Web interface
- Desktop application