Automatic Affect Classification of Human Motion Capture Sequences in the Valence-Arousal Model

Abstract

The problem that we are addressing is that of affect classification: analysing emotions given input data. There are two parts to this study. In the first part, to achieve better recognition and classification of human movement, we investigate that the labels on existing Motion Capture (MoCap) data are consistent with human perception within a reasonable extent. Specifically, we examine movement in terms of valence and arousal (emotion and energy). In part two, we present machine learning techniques for affect classification of human motion capture sequences in both categorical and continuous approaches. For the categorical approach, we evaluate the performance of Hidden Markov Models (HMM). For the continuous approach, we use stepwise linear regression models with the responses of participants from the first part as the ground truth labels for each movement.