Analysis of Computer Vision-Based Physiological Indicators for Operator Fatigue Detection

Detecting human mental fatigue is crucial because it significantly impacts work efficiency, especially in system operation control. In our earlier research, we developed a video-based system for detecting mental fatigue using physiological indicators analyzed through deep learning models. However, this approach is computationally expensive and requires processing of videos with various machine learning models to estimate multiple vital signs and facial features. To improve its efficiency, this paper introduces feature importance techniques that identify the most significant indicators of fatigue. This not only enhances our model’s performance but also provides insights into the factors contributing to fatigue. Additionally, we transitioned from using traditional machine learning techniques (i.e. Random Forest) to a more advanced architecture such as the Tabular Transformer. This shift offers several advantages, including better generalization and effective handling of tabular data, resulting in an impressive accuracy of 89%. Our findings promise to deliver more accurate and reliable assessments of mental fatigue in real-world applications.