Artificial Intelligence Applied to the Control and Monitoring of Construction Site Personnel

Abstract

Many countries are working towards gradually lifting restrictions generated by the COVID-19 virus as post-quarantine measures. The construction industry has had to adapt to new forms of work with economic and physical restrictions. For physical restrictions, the most worrying one is the risk of contagion, as many studies have indicated that social distancing is one of the most effective biosecurity measures. In this research, a training process was executed on a neural network to ensure an adequate social distance policy in a construction environment to identify people inside construction sites. More specific training was carried out to identify people performing activities in a position other than being completely upright, as is usually the case with construction workers. The “You Only Look Once” (YOLO) version 4 algorithm was used to train 2 classes of objects, an upright person and a crouched person. More than one thousand images of a construction site were used as a data set, achieving an accuracy of 77.98%. This research presents the results and recommendations to detect the people and calculate the distance between them. Based on the distance calculation, an alert report can be generated for the work areas for the health and safety team to take preventive actions.