A computer vision-based approach for Safety Management in the Construction Sites

Abstract

Personal protective equipment is one of the preventive methods to protect workers from harmful contacts. However, the utilization of this equipment is very poor in developing countries like Ethiopia due to shortage of safety material, lack of training, poor management, and improper utilization. Therefore, to overcome this challenge, local researchers recommend effective safety management. However, they do not include technological solutions on safety management that convert manual into automatic. Thus, this research aims to develop an easy monitoring system through computer vision to detect Personal protective equipment during construction work. Therefore, to achieve the study objective, ground+3 and above building projects are taken as a population, and the sample size is the ongoing projects which are 60 during the data collection period. Data was collected through questionnaires, observation, and interviews from the se lected construction companies and analyzed quantitatively and qualitatively to draw results and conclusions. To analyze data IBM SPSS statics and Microsoft Excel were used. This paper also presents a deep learning built on You-Only-Look-Once (YOLO) architecture to verify compli ance of workers’ safety clothes; if a worker is wearing a helmet, vest, safety shoes, and their combination from images, which is collected from the website and manually to test and train. Several experiments have been conducted, and the obtained detection accuracy 96.97% for hel mets, 97.73% for a reflector, and 98.9% for shoes. The obtained results have demonstrated the ability to detect Personal protective equipment with high precision (98%). The findings show that the vision-based method can be used for safety management to detect the presence and absence of personal protective equipment in a construction site.