Abstract
Construction projects’ unsatisfactory performance has been linked to factors
influencing individuals’ well-being and mental alertness on projects. Drowsiness is a
significant indicator of sleep deprivation and fatigue, so being able to identify the cognitive
and physical preparedness of workers on site to engage in construction tasks is important.
As a consequence of the strenuous nature of the work involved in construction, long work
hours, and environmental conditions, drowsiness is commonplace and has received less
attention despite being a leading cause of accidents occurring on-site. Detecting drowsiness
is essential for determining the safety and well-being of site workers. This study presents a
vision-based approach using an improved version of the You Only Look Once (YOLOv8)
algorithm for real-time drowsiness exposure among construction workers. The proposed
method leverages computer vision techniques to analyze facial and eye features, enabling
the early detection of signs of drowsiness, effectively preventing accidents, and enhancing
on-site safety. The model showed significant precision and efficiency in detecting
drowsiness from the given dataset, accomplishing a drowsiness class with a mean average
precision (mAP) of 92%. However, it also exhibited difficulties handling imbalanced
classes, particularly the underrepresented ‘Awake with PPE’ class, which was detected
with high precision but comparatively lower recall and mAP. This highlighted the necessity
of balanced datasets for optimal deep learning performance. The YOLOv8 model’s average
mAP of 78% in drowsiness detection compared favorably with other studies employing
different methodologies. The system improves productivity and reduces costs by preventing
accidents and enhancing worker safety. However, limitations, such as sensitivity to
lighting conditions and occlusions, must be addressed in future iterations.