Predicting Perceived Exhaustion in Rehabilitation Exercises Using Facial Action Units

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Human-Computer Interaction

Background:

• Physical exercise is crucial for managing non-communicable diseases, reducing mortality risk without medication.
• Artificial systems can enhance exercise adherence by monitoring progress and providing motivational feedback.
• Monitoring user exhaustion and offering tailored encouragement is vital for effective exercise coaching.

Purpose of the Study:

• To investigate the feasibility of predicting subjective exhaustion levels using non-invasive, non-wearable technology.
• To develop a system that monitors and provides feedback on exercise intensity based on physiological cues.

Main Methods:

• A novel dataset was created using facial recordings from 60 participants (30 male, 30 female).
• Seventeen facial action units (AU) were extracted as predictor variables.
• Subjective exhaustion was measured using the BORG scale, and regression/classification models were evaluated.

Main Results:

• Decision tree and support vector methods demonstrated reasonable accuracy in predicting exhaustion levels.
• Facial action units showed potential as indicators of perceived exertion.
• The study highlights the possibility of real-time exhaustion monitoring through facial analysis.

Conclusions:

• Non-invasive facial analysis can potentially predict subjective exhaustion during exercise.
• This technology could support personalized coaching and improve adherence in physical rehabilitation.
• Future research should address limitations such as participant bias and subjective facial expressions.