Development of person-independent emotion recognition system based on multiple physiological signals

An emotion recognition system based on processing of multiple physiological signals is presented. Our algorithm is developed and verified based on multiple subjects by presenting multimodal stimuli that were elaborated to effectively induce emotions, and thus not dependent on single specific user. The system utilizes physiological signals that can be acquired without discomfort from body surface. Support vector machine was introduced as pattern classifier to overcome the difficulty of large overlap among clusters and large variance within cluster. Correct classification ratio was 77.8 and 61.2%, for emotion recognition problem of three and four categories, respectively. Another advantage of our system is that it requires shorter signal monitoring time than previous ones, and thus better suited for practical use.