Personalized alertness prediction using video-based ocular and facial features.

Area of Science:

• Neuroscience
• Human Performance
• Biomedical Engineering

Background:

• The psychomotor vigilance test (PVT) is standard for assessing alertness impairment but is impractical for real-world use due to its time-consuming nature.
• Existing methods for alertness monitoring often require active participation, limiting their applicability in daily life.

Purpose of the Study:

• To develop a non-intrusive method for predicting individual alertness impairment using passively recorded facial videos.
• To establish personalized alertness-impairment predictions as a practical alternative to traditional PVT assessments.

Main Methods:

• Facial and ocular features were extracted from video recordings of 26 participants undergoing 62 hours of total sleep deprivation.
• Linear mixed-effects models were trained using these facial features to predict PVT performance (mean reaction times).
• Personalized alertness-impairment predictions were generated by integrating predicted PVT data into the unified model of performance (UMP).

Main Results:

• The models achieved an average root mean square error (RMSE) of 39 ms for predicting PVT mean reaction times from facial features.
• Personalized UMP predictions, based on video-derived PVT estimates, showed an average RMSE of 36 ms, closely matching the within-participant variability of rested conditions (30 ms).

Conclusions:

• A practical, proof-of-principle approach was developed to predict individual alertness impairment using passively recorded facial videos.
• This method offers a feasible and non-disruptive way to monitor and predict changes in alertness in real-world settings.