An incremental and interactive affective posture recognition system

Abstract

Date: 24-29 July 2005

Abstract: The role of body posture in affect recognition, and the importance of emotion in the development and support of intelligent and social behavior have been accepted and researched within several fields.
While posture is considered important, much research has focused on extracting emotion information from dance sequences. Instead, our focus is on creating an affective posture recognition system that incrementally learns to recognize and react to people's affective behaviors. In this paper, we examine a set of requirements for creating this system, and our proposed solutions. The first requirement is that the system is general and non-situation specific. Secondly, it should be able to handle explicit and implicit feedback. Finally, it must be able to incrementally learn the emotion categories without predefining them. We tested and compared the performance of our system using 182 standing postures described as a combination of form features and motion low level features, across several emotion categories, with a typical algorithm used for recognition, back-propagation, and with human observers in an aim to show the generalizability of the system. This initial testing showed positive results.