The Emoiton Correlated Response Patterns Of RR Interval And Skin Conductance

In the field of emotional human-computer interaction, the real-time emotion recognition from physiological signals will be one of the important applications of wearable intelligent, such as the dress that changes its color according to the owner’s mood or glasses that send out nice smell when the owner is feeling down. Compared to other physiological signals, such as EEG, facial EMG, or ECG etc., skin conductance (SC) and heart beat to beat interval (RR interval) signals have the advantages of easy measurement, strong anti-interference ability, and with minimal impact on user actions when in measurement, are the most suitable signals for online emotion recognition system. Therefore, it is of great value to analyze the emotion correlated response patterns of RR interval and SC.The aim of emotion correlated response pattern analysis is to extract the emotion-sensitive features of the physiological signals and to analyze the relationship between emtional states and the values of the features. Domestic and foreign researches on emotion recognition from physiological signals have bring a lot of inspiration for our work, however, there are still many problems to be solved, such as the various emotion models in the studies and lack of personalized analysis. In this paper, the problems in the processing of affective physiological signals and corresponding solutions are discussed, mainly including:1) For the problem of nonlinear analysis of physiological signals, propose an surrogated data based nonlinear feature validation method and two nonliner features, named subspace entropy and symbolized entropy, which are suitable for the analysis on short and noisy physiological signals.2) For the problem of data analysis with imprecise labels, propose a PSO-FCM clustering method and introduce a new criterion named cluster index to evaluate the clustering result.3) For the problem of user-independent emotion recognition, propose a baseline mapping method, which map the personal physiological baseline to the standard baseline.4) For the problem of personalized emotion recognition, propose a weighted method to adjust the recognition result. The experimental results suggest the effectiveness of these methods.The correlations between the changes of physiological signals, RR interval and SC, and the dimensions of emotion, valence and arousal, are discussed. The results show that the changes of RR interval and SC are correlated with emoitonal arousal and basicly uncorrelated with valence. Five features, RRmean, RRmin, RRLFVLF, SCnLF and SCmutual information are sensitive to emotional arousal. When the emoitonal arousal is very low, the mean RR interval and minimum RR interval are large (RRmean=0.83, RRmin=0.69), and the waveform of RR interval is smooth (RRLFVLF=2.37). The skin conductance level (SCL) changes gentle and decreases slowly, and fluctuates in a narrow range (SCnLF=0.90, SCmutual information=3.60). When the emoitonal arousal is low, the mean RR interval and minimum RR interval are medium (RRmean=0.80, RRmin=0.66), and there are short-term instability and slight fluctuations in RR intervals (RRLFVLF=0.99). SCL changes gentle and decreases slowly, and no high peak skin conductance response (SCR) appears (SCnLF=0.91, SCmutual information=3.67). When the emoitonal arousal is medium, the mean RR interval and minimum RR interval are medium (RRmean=0.80, RRmin=0.65), and there are short-term instability and slight fluctuations in RR intervals (RRLFVLF=1.08). SCL changes gentle and increases slowly, and no high peak SCR appears (SCnLF=0.83, SCmutual_information=3.41). When the emoitonal arousal is high, the mean RR interval and minimum RR interval are small (RRmean=0.77, RRmin=0.61), and there are short-term instability and slight fluctuations in RR intervals (RRLFVLF=0.99). SCL continues to increase, and high peak SCR appears (SCnLF=0.86, SCmutual_information=3.31). When the emoitonal arousal is very high, the mean RR interval and minimum RR interval are small (RRmean=0.75, RRmin=0.59), and there are long-term instability and slight fluctuations in RR intervals (RRLFVLF=0.93). The value of SCL is large, and high peak SCR frequently appears (SCnLF=0.77, SCmutual_information=3.30).This paper builds a prototype of the real-time personalized user-independent emotional arousal recognition system:1) In the aspect of user-independent, propose a baseline mapping method, which map the personal baseline to standard baseline that can remove the individual baseline differences. The cluster centers getting from the single-subject dataset are verified on multi-subject dataset. The result suggests the effectiveness of the cluster centers in recognizing emotional arousal on a user-independent way. The emotional arousal is quantitatively described by a value between 0 and 1, which is derived from the clustering result by arousal mapping.2) In the aspect of personalizd recognition, propose a weighted method to adjust the emotional arousal recognized from physiological signals. Discuss the impact of gender and personality on both emotinal arousals from self-feeling and recognized from physiological signals. The result is gender impacts more on recognized emotional arousal but less on self-feeling emotional arousal. However, personality impacts more on self-feeling emotional arousal but less on recognized emotional arousal. It suggests that the self-feeling emoitonal arousal is not always the same with recognized emotional aroual for one person. Therefore, a wighted method is proposed to adjust the recognized emotional arousal which enforces the consistency between self-feeling and recoginzed emoitonal arousal.Build the prototype of a real-time personalized user-independent emotional arousal recognition system. It contains the real-time RR interval and SC collection, the individual baseline measurement, the feature extraction, the clustering center setting, the emotional arousal recognition, the weighted adjustment, the personality factor setting and the emotional arousal display eight functional blocks.