PPE Detection for Construction Site Safety Leveraging YOLOv8

Abstract

Ensuring the safety of construction workers is paramount, with a critical aspect being the utilisation of Personal Protective Equipment. Annually, the construction industry alone contributes 24.2% of fatalities, averaging 38 people per day, out of an estimated 48,000 fatal occupational accidents occurring in India. Many accidents occur because of avoidance of safety measures, particularly the failure of workers to wear PPE. This study developed a novel framework utilizing YOLOv8, an up-to-the-minute deep learning model, to detect and ensure compliance with PPE protocols in real-time. The model is trained on a random dataset, available on Roboflow, consisting of images of workers equipped with PPE. The dataset includes labels of ’Hardhat’, ’Mask’, ’NO-Hardhat’, ’NO-Mask’, ’NO-Safety Vest’, ’Person’, ’Safety Cone’, ’Safety Vest’, ’machinery’, and ’vehicle’. This detailed annotation format, particularly distinguishing between the presence and absence of PPE, enhances its utility for tracking and monitoring applications. The dataset contains 10 classes with label annotations in YOLO format and includes metadata.csv and count.csv files for dataset information and train-val-test count. The model underwent training substituting a total duration of 2.719 hours. The experimental outcomes reveal the model’s exceptional accuracy of 76.3 percent and real-time capabilities, underscoring the effectiveness and practicality of computer vision methods in automating safety procedures at construction sites.

Keywords: - PPE detection, YOLOv8, construction safety, computer vision, deep learning.