Modeling Emotion Perception from Body Movements for Human-Machine Interactions Using Laban Movement Analysis

Emotion recognition from body movements is based in humans on the existence of associations in our brains between certain movements and specific emotions. These associations are responsible for the expression of emotions using typical movements, and for the elicitation of emotions by moving those typical associated movements. When observing other people move, our mirror neurons simulate in our brains the movements we see. This simulation elicits the emotion associated with the observed movements, enabling us to recognize it. To enable AI to recognize through computer vision the emotions expressed in movements, we have to provide the computer information about which movements are associated with which emotion. This chapter describes research which identified these associations, using Laban Movement Analysis (LMA) to portray the movements. We first demonstrated that both execution and observation of movements that express anger, happiness, sadness or fear, elicit the associated emotion. Next, we extracted from the movements used in the first experiment the LMA motor components that constructed those movements. Examining the effects of moving different combinations of those components, we identified which motor components are associated with, and enhance which emotion. We then established that movements composed of motor components associated with a specific emotion are perceived as expressing that emotion. Lastly, we demonstrated automatic recognition of these motor components using machine learning. Using LMA motor components (motor characteristics) to characterize emotional motor behavior rather than using a list of specific movements enables the identification of the associated emotion in any motor behavior.