Towards valence detection from EMG for Virtual Reality applications.

Abstract

The current practical restraints for facial expression recognition in Virtual Reality (VR) led to the development of a novel wearable interface called Faceteq. Our team designed a pilot feasibility study to explore the effect of spontaneous facial expressions on eight EMG sensors, incorporated on the Faceteq interface. Thirty-four participants took part in the study where they watched a sequence of video stimuli while self-rating their emotional state. After a specifically designed signal pre-processing, we aimed to classify the responses into three classes (negative, neutral, positive). A C-SVM classifier was cross-validated for each participant, reaching an out-of-sample average accuracy of 82.5%. These preliminary results have encouraged us to enlarge our dataset and incorporate data from different physiological signals to achieve automatic detection of combined arousal and valence states for VR applications.