Unobtrusive, Camera-Based Fatigue Detection During an Overnight Simulated Flight: A Two-Phase Calibration and Validation Study

Introduction: In shiftwork operations, including aviation, night work and other schedule demands can contribute to increased fatigue, which is associated with performance impairment. Standard fatigue assessments use either subjective, performance-based, or physiological measures, however, their utility can be hindered by a need for repeated assessments to track changes over time, report bias in the case of self-report measures, interference with work tasks, and other confounds and constraints. To help detect and manage fatigue, there is need for an unobtrusive system with continuous monitoring of fatigue that reliably tracks performance impairment. A camera-based, face- and eye- tracking system has been developed for this purpose (FOVIO Driver Monitoring System, Seeing Machines). An independent validation of this system was conducted against a standard, objective fatigue measure within a simulated nighttime flight operation setting.

Methods: The two-phase study used professional flight instructors, simulator technicians, local flight school students, and others able to act as captain or first officer for simulated flights in a high-fidelity Boeing 767 simulator. Participants were paired and a priori assigned to either the calibration or validation phase; the more experienced in each pair served as captain. They completed a 6-hour overnight flight (00:00–06:00) with continuous recording using the camera-based system. Concurrently, objective performance measurements were taken using a 10-minute psychomotor vigilance test (PVT) at 6 time points across the night corresponding with key times within a flight. The camera-based system used an infrared camera with two IR illuminators mounted in front of each participant and Linux-based software to continuously record and process the data to generate drowsiness scores. For the calibration phase, the camera-based data were fitted to PVT mean response times (mean RT) using mixed-effects regression analysis. Goodness-of-fit between PVT means RTs and the camera-based predictions were quantified with the root-mean-square-error (RMSE). For the validation phase, the regression coefficients from the calibration phase were retained, allowing camera-based drowsiness levels to be linearly transformed to PVT mean RT predictions without further model fitting (i.e., no re-calibration occurred). Again, goodness-of-fit was quantified with the RMSE. A bootstrap analysis was then used to statistically compare the calibration and validation RMSEs.

Results: Data were available for 14 individuals in the calibration phase and 12 individuals in the validation phase (ages 20–50, 3 females). The RMSE for the calibration phase was 68.2ms, reflecting acceptable prediction accuracy. The RMSE for the validation phase was 82.0ms, which was not significantly different from the calibration phase (p = 0.219) and indicated that acceptable prediction accuracy was retained.

Conclusion: The findings of this validation study constitute independent, scientific evidence of the adequacy of this camera-based system to measure objective fatigue in a simulated nighttime flight operation. Further research is needed to verify that these findings generalize real-world flight operations. Support: Research supported by Federal Express Corporation, camera-based system provided by Seeing Machines (neither company was involved in data collection, analysis, reporting).

Publikationsverlauf

Artikel online veröffentlicht:
08. Oktober 2025