Multiclass SVM for Affect Recognition with Hardware Implementation

摘要

Emotion recognition is an interested field in the affective computing domain thanks to its important applications in Human Machine Interaction. Recent investigated studies proved the effectiveness and the relevance of physiological response as a modality to detect the human emotional states. This paper deals with an improved emotion recognition system that aims to classify the affective states into three defined areas in Arousal-Valence Space. Basing on previous works, we explored only two physiological signals, namely, electrocardiogram and respiration amplitude which were collected in the publicly available database MAHNOB-HCI. After preprocessing the signals, we extracted and normalized emotionally relevant features from these two signals. Prior to the classification stage using multiclass support vector machine as a classifier, we applied a level feature fusion. Furthermore, we implemented the classification stage on the Raspberry Pi III Model B using Python platform. In this work, we achieved 60.41 % for Arousal and 59.57% for Valence. The obtained classification rates are promising compared to recent related works.