earticle

영어

Emotion recognition technology, a core for human-robot interaction, must have fast data processing speed and real-time characteristics. This paper proposes an emotion classification method using two bio-signals for real-time emotion recognition. The proposed method is a technique to classify human emotions into three (high or low, neutral arousal / valence) emotions by extracting features from one pulse of PPG(Photoplethysmogram) signal and GSR(Galvanic Skin Response) signal through a 1D CNN model. For real-time emotion recognition, input data requires a short length from 1 to 3 seconds and fast preprocessing. Therefore, in this paper, we use three preprocessing processes and one short pulse of Window size 1.1s without feature extraction, and the label of each pulse uses annotation labeling. The Experiments show that average accuracy is 71.1% arousal and 71.8% valence, confirming the possibility of emotion recognition with pulse-wise data and a real-time possibility with 0.17 seconds of preprocessing time per sample data.