Poincaré's section analysis for PPG-based automatic emotion recognition

Given the importance and abundant use of emotional recognition in the human–computer interfaces and multimedia applications, several methods for designing such systems have been reported in the literature. However, the performance of the photoplethysmography (PPG) signal in these devices has not been sufficiently studied. In addition, details of the phase space geometry of this signal have not been investigated in the emotional states up to now. In connection with this issue, Poincaré's sections can quantify the geometric patterns of the trajectory in the high-dimensional phase space. Therefore, this work attempted to propose an automatic emotion recognizer which can characterize the PPG signals during the exposure of emotional music-video. Our focus was to detect and classify dynamical behaviors of the PPG trajectories in three emotional classes of love, hate, and fun using Poincaré's section measures. To this effect, the 2D phase space of PPG was firstly reconstructed. Then, forming the Poincaré's sections in different angles, some geometric indices were extracted. Finally, using support vector machine, the PPG measures were classified into emotional states. Our results showed that basin geometry of the PPG phase states was significantly different in different emotional states. The maximum accuracy rates of 96.67% and 91.11% were achieved for a binary and multi-class classification scheme, respectively. The proposed framework was fast and operated using the single-sensor signal. In conclusion, the dynamical Poincaré's section indices of PPG signals during three emotional states paved the way for designing an online emotion recognition system.