Towards a Contactless Stress Classification Using Thermal Imaging

Area of Science:

• Physiological monitoring
• Biomedical engineering
• Human-computer interaction

Background:

• Thermal cameras non-invasively measure infrared radiation, reflecting autonomic nervous system (ANS) activity via skin temperature.
• Skin temperature variations are linked to physiological stress responses.
• Contactless monitoring is desirable for unobtrusive stress assessment.

Purpose of the Study:

• To evaluate the effectiveness of thermal imaging in automatic stress detection.
• To develop a contactless stress recognition system using thermal data.
• To compare the contribution of thermal features versus traditional ANS correlates in stress detection.

Main Methods:

• Recorded thermal images and ANS correlates (cardiac, electrodermal, respiratory activity) from 25 volunteers under induced acute stress (Stroop test).
• Extracted features from both thermal and ANS signals.
• Implemented subject-independent stress classification using Support Vector Machine (SVM) with recursive feature elimination.

Main Results:

• Thermal features significantly contributed to stress detection accuracy.
• A combined model of ANS correlates and thermal features achieved 97.37% accuracy.
• A model using only thermal features achieved 86.84% accuracy for contactless stress detection.

Conclusions:

• Thermal imaging is a valuable tool for contactless, automatic stress detection.
• Integrating thermal data enhances the accuracy of stress recognition systems.
• Contactless thermal monitoring shows significant potential for real-world stress assessment applications.