Natural Interaction expressivity modeling and analysis

G. Caridakis, K. Moutselos, I. Maglogiannis 
In Proc of 6th ACM International Conference on Pervasive Technologies Related to Assistive Environments (PETRA 2013), Rhodes, Greece ACM

Abstract: Behavior, including non verbal, expressiveness is an integral part of the communication process since it can provide information on the emotional state and the user’s performance when the aim of the interaction is measurable. Long term temporal measurements can also assist in monitoring the user for either emergencies or long term mood instabilities. Current article presents research work on the computational formalization and analysis of full body 3D expressivity in Natural (bodily) Interaction within the framework of Pervasive Assistance. Expressivity dimensions are selected as the most complete approach to body expressivity modeling, since they cover the entire spectrum of expressivity parameters related to emotion and affect. In this study five expressivity parameters are computationally formalized, using different approaches based on silhouette, limbs position and joints rotation, for each expressivity feature. These approaches are then evaluated in terms of their effectiveness in modeling the expressivity aspect in question. The modeling effectiveness of each approach is assessed using Linear Discriminant Analysis (LDA) and its coefficients on the automatically extracted parameters, defined in the computational formalization, against an experimental dataset consisting of extreme expressions (positive and negative) of the investigated expressivity aspects. The The experimental results confirm that the proposed Fading Silhouette Motion Volumes (FMSV) approach, is the most effective in modeling body expressivity.