Generalizable Gaze Synthesis for Practical Fatigue Detection Systems

Fatigue has been established as a significant factor affecting both productivity and safety across a range of industries. Traditional self-reporting methods are commonly used for fatigue detection, but they are inherently subjective and lack real-time monitoring capabilities. Although objective methods, such as electroencephalograms and facial features, have proven to be reliable and effective in detecting fatigue, their susceptibility to noisy environments and privacy concerns significantly constrain their practical applications. To address these issues, this paper proposes the VisioPhysio Fatigue Detector (VPFD) as a practical solution to fatigue management. By utilizing two commercially-available wearable devices: HoloLens 2 and Google Pixel Watch 2, VPFD integrates visual and physiological signals to detect fatigue in a non-intrusive. Given the limited publicly available datasets for fatigue and unique strengths of the modalities used in these datasets, the paper introduces an innovative data synthesis method that has been experimentally proven to be effective.