Personal protective equipment detection using deep learning

• Main Author: Qin, Bohao
• Other Authors: Yap Kim Hui
• Format: Thesis-Master by Coursework
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science; Engineering; Deep learning; YOLO; Personal protective equipment; Detection
• Online Access: https://hdl.handle.net/10356/174883
• Institution: Nanyang Technological University

Industrial safety management aims to ensure the safety of workers and prevent injuries. Personal protective equipment (PPE) is one of the most important safety measures that the field applies to protect construction workers from hazards. Therefore, it is crucial to ensure that workers are wearing assigned PPE and doing so properly in order to effectively minimize workplace accidents. Using deep learning techniques, we set out to develop AI models to automate detection of safety failures in PPE use. To achieve this, we utilize a deep learning-based object detector to detect PPEs and identify if workers are wearing them properly. Specifically, we review the literature on different off-theshelf object detector models, comparing their accuracy and efficiency and finding YOLOv8 exhibited the highest balance of performance among the assessed detectors in terms of both precision and efficiency. We further train YOLOv8 on a PPE detection dataset and experimentally compare it to a similarly trained YOLOv7 model using a suitable set of benchmarks. Our extensive evaluation results on the PPE dataset demonstrate that YOLOv8 outperforms the alternatives in terms of training speed, and detection efficiency, reaching a remarkable overall 86.5% mAP@0.5.