Inspiration

Mining and industrial workplaces are among the most hazardous environments, where workers face risks from missing Personal Protective Equipment (PPE), fire outbreaks, and smoke hazards. Manual monitoring is often inconsistent and delayed, increasing the chances of workplace accidents.

We wanted to build an AI-powered solution that continuously monitors safety compliance and detects hazards in real time, helping organizations prevent accidents before they occur.

What it does

MineShield AI is an intelligent industrial safety monitoring platform that uses computer vision to detect:

• PPE compliance (Helmet, Gloves, Vest, Boots, Goggles, Face Shield)
• Missing PPE violations
• Fire detection
• Smoke detection

The system supports multiple input sources including images, videos, webcams, and RTSP/IP camera streams. It performs real-time inference, highlights detected hazards with bounding boxes, and can save images, videos, labels, and JSON reports for further analysis.

How we built it

We trained custom YOLO11 models using curated datasets for PPE detection and fire & smoke detection.

The application is built using C++ and Python with OpenCV DNN for high-performance inference. We implemented image preprocessing using Letterbox resizing, confidence filtering, Non-Maximum Suppression (NMS), and GPU acceleration through CUDA.

The backend is designed to support multiple AI models running together while maintaining modularity for future expansion.

Challenges we ran into

• Collecting and merging multiple datasets while maintaining consistent class labels.
• Training accurate models for multiple PPE classes and hazard detection.
• Optimizing real-time inference performance for high-resolution video streams.
• Managing simultaneous execution of multiple detection models efficiently.
• Ensuring reliable detection under different lighting conditions, camera angles, and worker movements.

Accomplishments that we're proud of

• Successfully trained custom YOLO11 models for PPE, fire, and smoke detection.
• Built a modular AI pipeline capable of processing images, videos, webcams, and RTSP streams.
• Integrated GPU acceleration for faster inference.
• Designed a scalable architecture that can be extended with additional industrial safety modules.

What we learned

Throughout this project we gained hands-on experience in:

• Computer Vision
• Deep Learning
• YOLO11 training and optimization
• OpenCV DNN
• CUDA acceleration
• Real-time AI inference
• Industrial AI application design
• Dataset preparation and model evaluation

What's next for MineShield AI

We plan to extend the platform with:

• Worker tracking
• Restricted zone intrusion detection
• Fall detection
• Emergency alert notifications
• Dashboard analytics
• Cloud deployment
• Edge AI optimization using TensorRT

Built With

• artificial
• c++
• cmake
• computer
• cuda
• deep
• fastapi
• git
• github
• intelligence
• learning
• machine
• onnx
• opencv
• python
• react
• sqlite
• vision
• yolo
• yolo11