Deep Learning based Emotion Recognition through Biosensor Observations

Emotion is the conscious experience which can be characterized by functional mental activity and by the degree of contentment and discontentment. In this paper, we have studied physiological changes to infer emotions from biosensor observations. We have considered four basic emotions i.e. Joy, Sad, Surprise and Disgust. The emotions are stimulated through video stimuli. The Likert scale is used to collect subject’s aerosol and valence level to determine the ground truth emotions of the user while watching and listening video stimuli. The wearable ECG, GSR and BVP sensor observations are collected from human subject to infer the emotion which is internally exposed through video stimuli. The convolutional neural network (CNN) of deep learning architecture is used to infer emotions from biosensor’s signal features. The simulation results show higher accuracy of the proposed CNN based emotion recognition approach.