Personal Protective Equipment Detection Using Computer Vision Techniques

The intensive use of chemicals, machines and electrical assets introduced unsafe conditions to the workplace. An unsafe condition is a physical condition that can cause an incident, such as operating without training, defective supplies and poor housekeeping. Such conditions might cause serious injury or even death. As well as the human impact, unsafe conditions have a significant impact on operational excellence and the financial state of a company. Companies are committed to ensure a safe environment by setting safety polices, conducting safety training, fire prevention systems, safety manuals and signboards and providing safety gears. Personal protective equipment (PPE) is safety equipment that can maintain the safety of employees in hazardous conditions, such as hot surfaces and toxic chemicals that can cause serious injuries and illness. PPE is sometimes referred to as the last line of defense. Some workers might not comply with safety policies or refuse to wear the PPE. To overcome the manual safety checks and compliance of employees, in this paper we propose an AI-powered computer vision automation solution leveraging the state of the object detection model. Computer vision is the field that mimics human vision to extract purposeful information from videos and images. Computer vision brings about various functionalities to perform tasks such as object detection, object classification, object identification and object verification. The proposed solution is developed by using a computer vision technique that detects various types of PPEs in real time. The main purpose of this project is to detect a presence of eight classes (person, helmet color: Red, Yellow, Blue and White, head, vest, glasses). The best results are achieved by applying YOLOv5 on a set of construction site images with corresponding annotations in YOLO format. The proposed solution automates the process of detection and monitoring PPE and employee behavior in operation fields in real-time. Automating the detection can reflect the business value by reducing the timeframe for tracking, creating a safe environment that in turn can increase the productivity and safety of the workers and reduce the costs of operations. The proposed solution includes all the components of data ingestion, data processing, object detection model and deployment on the edge device or server to improve safety.