A Preliminary Approach to Identify Arousal and Valence Using Remote Photoplethysmography

This study presents a preliminary evaluation of using features linked to the Heart Rate Variability (HRV) extracted from the remote photoplethysmography (rPPG) signal to identify emotional levels (low, neutral and high) of arousal and valence. The method for obtaining the rPPG is a modified version of the sub-band algorithm and the Plane Orthogonal to Skin (POS) techniques. The emotion classification was evaluated in two forms: (1) personalized (or intra-subject); and (2) cross-subjects (or inter-subjects). The emotion-classification analysis were performed with feature vectors from HRV-based rPPG, and compared with the result from two traditional methods: the photoplethysmography (PPG) and the electrocardiography (ECG). The classification accuracy results indicate that: the proposed method cannot discriminate the arousal of the subjects, however for valence, the classifier got better results when rPPG is used, with accuracy of 51 % for personalized classification and 42 % for cross-subjects.