Comparison of YOLO Algorithms for Sitting Posture Assessment of Office-Based Workers in Construction

Abstract

Improper sitting posture is a significantly contributes to physical and psychological damages to office-based workers particularly where there is a lack of continuous ergonomic monitoring. This study analyses four automated sitting posture detection models based on recent versions of YOLO object detection algorithm. The models demonstrated reliable sitting posture classification by accurately detecting sitting posture variation. The proposed models provided an efficient solution for office-based workers siting posture monitoring compared to existing sensor based and computer vision-based models. The models were evaluated based on key performance matrices such as true positive values, precision score, recall score, mAP values and inference speed. Among the four models, YOLOv11-s was selected as the optimal model due to its performance in accuracy and computational efficiency. The model can be further developed by integrating real-time feedback mechanisms, such as automated posture alerts to improve workplace ergonomics, worker wellbeing, and overall safety practices.