Emotional stress detection using facial thermal imaging

Abstract

Thermal imaging (TI) is a noninvasive method for assessing psychological states. This paper examines the features of physiological responses to emotional stress (ES) and their detection by TI. The correlations between thermal patterns and stress markers (cortical level and heart rate [HR]) are weak and not statistically significant, suggesting limited direct association. However, facial thermal patterns show sensitivity to ES and could allow real-time stress detection. Temperature immediately rises during startle responses, but experimental results indicate that the periorbital region’s temperature does not rise immediately during the Trier Social Stress Test but instead shows a delayed increase after stress begins. Moreover, this paper develops periorbital thermal signal-based HR extraction using a novel region of interest to estimate HR in real-world contexts. Further, this paper proposes Diverse Energy-Phase Space Attention Magnification Network to amplify and extract stress-induced signals. The encouraging results demonstrate 97.1% accuracy in ES detection. Overall, these experimental results exhibit that facial TI is a valuable tool for estimating ES.