Research On Emotion Recognition Method Based On Multi Physiological Signals

With the development of computers, people wish to give computer emotional intelligence and the ability to interact with humans emotionally. Giving computers the ability to recognize human emotion accurately is key technology to achieve the vision. In previous studies, emotions were often recognized based on human face expression, voice, posture or gestures. However, as physio logical signals are also correlated with human emotion and can hardly be masked, recognizing emotions based on physiological signals has become a research hotspot in lecent.This paper presents a method of recognizing four basic emotions (calm, joy, sadness, and fear) based on four physiological signals (electrocardiogram, respiration, pulse wave and skin conductance).17 subjects’physiological signals in an affective state of calmness, joy, sadness and fear were collected and a total of 668 samples were obtained. The time domain and frequency domain characteristics of the samples were analyzed and signal features were extracted. Aiming at the problem of overhigh dimension of the raw feature space, a feature selection method based on a Sequential Forward Floating Selection (SFFS) algorithm was applied, which selects a set of key features from the raw features after feature quantity is specified. Based on the selected key features, a support vector machine (SVM) algorithm was utilized to train and predict the above four emotions. To solve the problem that the performance of SVM is influenced by penalty factor and kernel parameters significantly, a genetic algorithm was utilized to optimize the penalty factor and the kernel parameters. The performance of the emotion recognition method was evaluated using cross validation and an accuracy of 71.4% was achieved with 31 key features. Besides, the algorithm was applied on a physiological signal database provided by Augsburg University and an accuracy of 95% was achieved. The results show that the presented method can recognize human emotions effectively and has relatively good universality.