A Study Of Affective Computing Based On Multi-physiological Signal

Affective computing is becoming increasingly important with the development of human-machine interaction technology.Because of the different ways of emotional expression,large numbers of researches have already been made,including voice,facial expressions,and gestures.But the research on the connection between autonomic nervous system(ANS)response and emotion is insufficient.ANS response cannot be changed intentionally or unintentionally as well as other emotion expression,because ANS response,such as heart rate,breathing reaction,is produced by the activity of ANS activity which laid the foundation for the study of affective computing based on multi-physiological signals.In this paper,we mainly study the application of ECG,RSP and EDA signals in affective computing,and discuss the whole process from signal collection to emotion model.In order to collect available signals,we successful design a database which based on the reaction of dynamic pictures adds music in multiperson by different arousal and valence level.Since using physiological signal to express emotion is not intuitive and proving affective computing based on physiological signals is hard,this paper proposes a concept named biomusic which encode physiological signals to music using Musical Instrument Digital Interface(MIDI)and design an experiment proving the validity of biomusic.The result of experiment has shown that biomusic is effective in affective computing and fusion of multi-physiological signals.After demonstrating the effectiveness of affective computing,this paper proposes an efficient model for affective computing from the aspects of signal feature extraction,feature preprocessing,feature selection and emotion modeling.Based on the characteristics of each signal,we extract significant features for each signal.To ensure the effectiveness of the features,we use the random forest algorithm as a feature selection method which extracts features according to the importance of each feature.In order to verify the selection of feature,we using linear discriminant analysis(LDA)as affective computing model and analyze the result of valence and arousal in a visual way.The experimental results show that using random forest and linear discriminant analysis model is more efficient than other models,and the recognition rate of valence and arousal is relatively high.The results of this paper provide a reference for future research and experience.