Feasibility Analysis of PPE Detection Via YOLOv7 & YOLOv7-Pose to Enhance Workplace Safety

Ensuring workplace safety in Pre-Fabrication environments is important, particularly in adhering to safety compliance protocols through the correct usage of Personal Protective Equipment (PPE). However, existing systems often fall short in detecting whether PPE is correctly worn or not, posing a risk to workplace safety. By creating a specialized dataset for Pre-Fabrication environments, the system can detect four PPE tools which are hard hat, glove, safety shoes and welding mask. The dataset contains 934 training images and 255 images for testing and validation, respectively. After rigorous training, testing and validation process, the experimental result and performance evaluation show that the PPE detection model achieved 96.8% precision, 95% recall and a mean Average Precision (mAP) of 0.879. However, with the current model, there are several false positive errors from the result where the model falsely detects other objects as PPE. Therefore, this study proposes a dual-model system employing YOLOv7 object detection and pose estimation models to further improve the accuracy of the PPE detection model and reduce false positives. This dual-model system, validated through live CCTV footage, provides a comprehensive solution for enhancing safety protocols in Pre-Fabrication workspaces.