Machine learning-based recognition of mental fatigue in construction equipment operators using facial features.

Construction equipment operators are at risk of mental fatigue, which can lead to accidents and health problems. Real-time monitoring is necessary to prevent accidents and protect operators' well-being. Previous studies have used wearable sensors to classify mental fatigue in operators, but these methods require physical sensors to be worn, causing discomfort and irritation. Therefore, a new approach is needed that allows for contactless measurements of mental fatigue. In this study, a novel approach was proposed using machine learning and geometric measurement of facial features to classify mental fatigue states during equipment operations. Video recordings were obtained during a one-hour excavation operation, and four facial features (eye distance, eye aspect ratio, head motion, and mouth aspect ratio) were extracted for analysis. The temporal increase in NASA-TLX score was used as the ground truth for mental fatigue. The results showed that the support vector machine classifier outperformed, achieving a high accuracy of 91.10% and an F1 score between 85.29% and 95.61%. These findings suggest that mental fatigue in construction equipment operators can be non-invasively monitored using geometric measurements of facial features.