Multimodal analysis of eye movements and fatigue in a simulated glass cockpit environment

Abstract

Pilot fatigue is a critical reason for aviation accidents related to human errors. Human-related accidents might be reduced if the pilots’ eye movement measures can be leveraged to predict fatigue. Eye tracking can be a non-intrusive viable approach that does not require the pilots to pause their current task, and the device does not need to be in direct contact with the pilots. In this study, the positive or negative correlations among the psychomotor vigilance test (PVT) measures (i.e., reaction times, number of false alarms, and number of lapses) and eye movement measures (i.e., pupil size, eye fixation number, eye fixation duration, visual entropy) were investigated. Then, fatigue predictive models were developed to predict fatigue using eye movement measures identified through forward and backward stepwise regressions. The proposed approach was implemented in a simulated short-haul multiphase flight mission involving novice and expert pilots. The results showed that the correlations among the measures were different based on expertise (i.e., novices vs. experts); thus, two predictive models were developed accordingly. In addition, the results from the regressions showed that either a single or a subset of the eye movement measures might be sufficient to predict fatigue. The results show the promise of using non-intrusive eye movements as an indicator for fatigue prediction and provides a foundation that can lead us closer to developing a near real-time warning system to prevent critical accidents.