QHSE Object Detection for Workplace Safety Monitoring

Project Overview

QHSE Object Detection

QHSE Object Detection is a computer vision-based prototype designed to simulate workplace safety monitoring scenarios. This dummy project adapts core concepts from a real-world internship, focusing entirely on building and benchmarking object detection models for various Quality, Health, Safety, and Environment (QHSE) applications.

All models were trained and evaluated using synthetic and publicly available datasets from Roboflow, Google, and other open data sources. No confidential or production data was used in this project.

Core Features

• PPE Compliance (No Helmet Detection): Detects workers not wearing helmets in simulated container yard environments.
• Fire Detection: Identifies visible flames as early indicators of hazards in storage or operational areas.
• Drowsy Driver Detection: Classifies behaviors such as eye closure and yawning to identify potential driver fatigue.
• Seatbelt Driver Detection: Detects whether drivers are wearing seatbelts (implemented using YOLOv11s only).
• Fall Detection via Pose Estimation: Compare YOLOv11s and MediaPipe to estimate human poses and identify fall incidents in high-risk or slippery zones.

Model Results and Analysis

A comparison was conducted between YOLOv11s and RT-DETR-L on three key tasks: No Helmet Detection, Fire Detection, and Drowsy Driver Detection.

• RT-DETR-L: Demonstrated higher detection accuracy than YOLOv11s.
• YOLOv11s: Slightly less accurate but faster and lighter, ideal for edge deployment.

Project Focus

This is a dummy prototype project created purely for research and educational purposes. It does not use any proprietary, sensitive, or company-specific data.

The main objective of this project is to explore the development, benchmarking, and performance evaluation of AI-based object detection models for potential use in QHSE-related applications. No backend system, real-time deployment, or hardware integration was implemented.

⚠️ Disclaimer: All dataset images are for non-commercial, educational use only. Credits belong to the original owners. Contact the author for removal or proper attribution.