Automatic Detection of Acute Mental Stress with Camera-based Photoplethysmography

Autonomic nervous system (ANS) activity reflects in vital signs that can be measured by means of camera-based photoplethysmography (cbPPG). This work investigates the automatic detection of acute mental stress with cbPPG. Data from the Dresden Multimodal Biosignal Dataset for the Mannheim Multicomponent Stress Test (DMBD-MMST) covering > 40 h uncompressed facial RGB videos of 56 healthy participants were used for rest vs. stress classification on the basis of nine cbPPG vital signs with decision tree ensembles. Also, the impact of normalization, measurement duration, and color channel combination was investigated. Best performance for rest (baseline and recovery) vs. stress classification ( $F1=0.81$ , Cohen's Kappa $\kappa=0.61$ ) was achieved with normalization, 30 s measurement duration, and vital signs from the green channel and the color channel combination called 03C. Without recovery (baseline vs. stress), this configuration achieved $F1=0.97$ and $\kappa=0.94$ . Paired t-tests revealed significant changes from rest (baseline and recovery) to stress in eight of the nine vital signs and the maximum effect size amounted to $d\ =\ 0.73$ , indicating sympathetic excitation. Findings from this work are central to the non-contact evaluation of ANS activity. Our results demonstrate that automatic detection of acute mental stress with cbPPG is feasible.