Development Of A Real-Time Monitoring System For Detect The Use Of Personal Protective Equipment (PPE) From Machine Learning

Resumo

PPE is the equipment for individual use used by the worker where its main purpose is to protect against
risks capable of jeopardizing their health and safety, in addition also reducing costs to the employer with personnel
replacements, dismissals and indemnity processes. However, many times, either through negligence or discomfort,
there is resistance to its use and/or the removal of the equipment during the performance of activities. In view
of this problem, the HSE department must therefore inspect and monitor the proper use of workers‘ personal
protective equipment mostly of the time. As an alternative to assist the security department in verifying, demand
and quantifying the use of protective equipment in the workplace, this project presents a machine learning model
based on the You-Only-Look-Once (YOLO) architecture to verify the workers‘ compliance regarding their safety
behavior in real time, using images / video of a security system installed in a busy place within an industry. The
algorithm uses the approach of detecting workers and basic PPE as helmet, gloves, goggles simultaneously by deep
learning previously trained by a image dataset of workers in differents types of labour ambient, and next verifies
that each bounding box generated is in the correct position, thus confirming if the worker is carrying PPE or not.
Later, a program developed in python using the opencv library will quantify the use or not of the use of PPEs,
from the bounding boxes generated by more specifically YOLOv4, providing in this way a statistical report as an
output. This statistical report will be useful for the HSE (Health, Safety and Environment) department to use its
statistics as indicators of reliability that will assist in decision-making and in the management of security within
the company.