Learning Long-Term Behavior through Continuous Emotion Estimation

Abstract

Emotions convey concise information regarding an individual’s internal state, while in the long-term they can be used to form an opinion about his/her overall personality. The latter can be proved particularly vital in many human-robot interaction tasks, like in the case of an assisted living robotic agent, where the human’s mood may in turn require the adaptation of the robot’s behavior. As a result, the paper at hand proposes a novel approach enabling an artificial agent to conceive and gradually learn the personality of a human, by tracking his emotional variations throughout their interaction time. To achieve that, the facial landmarks of the subject are extracted and fed into a Deep Neural Network architecture that estimates the two coefficients of human emotions, viz., arousal and valence, as introduced by the broadly known Russell’s model. Finally, by creating a dashboard for user-friendly display of our results, we present both momentarily and in the long-term the monitored fluctuations of a person’s emotional state.