Automatic Recognition of Experienced Emotional State from Body Movement
Published in International Conference on Human Computer Interaction, 2021
Voigt-Antons, J.-N., Devaikin, P. & Kojić, T.
Although body movement carries information about a person’s emotional state, this modality is not widely used in automatic emotion measurement systems. With this paper, we address the question of the automatic recognition of a person’s affective state by analyzing the way a person moves. We present the approach which was used to build a classifier of experienced, non-acted emotions. To collect the data to train and validate the classifier, a controlled laboratory study was conducted. During the study, the music mood induction procedure was used to evoke different emotions in the participants. The participant’s movement was recorded using a depth sensor, two accelerometers, and two electromyography sensors. An accuracy of 43% was achieved to recognize four emotion classes, corresponding to four quadrants of the valence-arousal space. For the participants with lower dance or movement proficiency, recognition was more accurate. The same was discovered for the participants with higher comfort levels of moving in front of a camera. The findings show the potential for the automatic analysis of a person’s body movements to gather information about an affective state.
Recommended citation: Voigt-Antons, J.-N., Devaikin, P. & Kojić, T. (2021, July). Automatic Recognition of Experienced Emotional State from Body Movement. Paper presented at the International Conference on Human Computer Interaction. Washington DC, USA. https://doi.org/10.1007/978-3-030-78462-1_49