Automatic Human Stress Detection Based on Webcam Photoplethysmographic Signals

One of the goals of affective computing field is to provide to computers the ability to recognize automatically the affective state of the user in order to have more intuitive human–machine communication. This paper aims to detect automatically the stress user when he is interacting with computer. The developed system is based on instantaneous pulse rate (PR) signal extracted from imaging photoplethysmography (PPG). Seven features from time and frequency domain are extracted from PR signal and processed by learning pattern recognition systems. Two methods based on Support Vector Machine (SVM) and Linear Discriminant Analysis (LDA) are used and compared to classify the user’s emotional state. A computer application based on “Stroop color word Test” is developed to elicit emotional stress in the subject. The proposed method can achieve the overall average classification accuracy of 94.42% and 91.10% with SVM and LDA, respectively. Current results indicate that our approach is effective for stress classification.