Identifying stress scores from gait biometrics captured using a camera: A cross-sectional study

Area of Science:

• Biomedical Engineering
• Computer Science
• Psychology

Background:

• Stress is a significant health risk factor lacking objective, non-intrusive measurement methods.
• Human gait, a pattern of locomotion, shows potential as a behavioral indicator for mental states.

Purpose of the Study:

• To develop and validate an objective, non-intrusive method for assessing stress severity using gait analysis.
• To identify key gait features indicative of stress levels.

Main Methods:

• 152 participants' perceived stress was measured using the Perceived Stress Scale (PSS-10).
• 2D video recordings of natural walking were analyzed to extract 1320 time-domain and 1152 frequency-domain gait features.
• Machine learning regression models (Gaussian Process, Linear Regression, Random Forest, Support Vector) were trained using the top 40 gait features.

Main Results:

• Models combining time- and frequency-domain gait features achieved the best performance (RMSE=4.972, r=0.533).
• Gaussian Process Regressor and Linear Regression models demonstrated superior accuracy in stress level assessment.
• Gait features from the waist, hands, and legs were identified as the most significant contributors to stress detection.

Conclusions:

• Machine learning models utilizing 2D video-based gait data can accurately detect stress severity.
• The developed approach offers a reliable, non-intrusive method for assessing perceived stress.
• Specific body movements, including those of the waist, hands, and legs, are crucial indicators for stress detection.