Pattern Recognition Algorithms for Gait Analysis with Application to Affective Computing

Abstract

This thesis studies the integration of nonverbal communication, here emotions, to human-robot interaction. Emotions are conveyed via different modalities. Within this context, the human gait is investigated as modality at distance. For this purpose, machine learning algorithms are developed to estimate the quality to recognize emotions in gait patterns. These results are interpreted in relation to human perception. Furthermore, it is studied to what extend the gait is applicable to express emotions in robotics using the example of a hexapod. In connection to the interdisciplinary direction of the project, mathematical relations between inferential and predictive statistics are elaborated. Finally, this work concludes that emotions are recognized in gait about chance level, even though the gait is a highly individual motion pattern.