Wearable Emotion Recognition System based on GSR and PPG Signals

In recent years, many methods and systems for automated recognition of human emotional states were proposed. Most of them are trying to recognize emotions based on physiological signals such as galvanic skin response (GSR), electrocardiogram (ECG), electroencephalogram (EEG), electromyogram (EMG), photoplethysmogram (PPG), respiration, skin temperature etc. Measuring all these signals is quite impractical for real-life use and in this research, we decided to acquire and analyse only GSR and PPG signals because of its suitability for implementation on a simple wearable device that can collect signals from a person without compromising comfort and privacy. For this purpose, we used the lightweight, small and compact Shimmer3 sensor. We developed complete application with database storage to elicit participant»s emotions using pictures from the Geneva affective picture database (GAPED) database. In the post-processing process, we used typical statistical parameters and power spectral density (PSD) as features and support vector machine (SVM) and k-nearest neighbours (KNN) as classifiers. We built single-user and multi-user emotion classification models to compare the results. As expected, we got better average accuracies on a single-user model than on the multi-user model. Our results also show that a single-user based emotion detection model could potentially be used in real-life scenario considering environments conditions.