Emotion Recognition Using Electrocardiogram and Photoplethysmorgraphy

• Main Authors: Meng-Yu Shih, 示孟玉
• Other Authors: Sung-Nien Yu
• Format: Others
• Language: zh-TW
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/51618578625321951636

碩士 === 國立中正大學 === 電機工程研究所 === 101 === In this thesis, we proposed an emotion recognition system based on physiological signals. Electrocardiogram (ECG) and photoplethysmorgraphy (PPG) were used to recognize seven kinds of emotions, including neutral (non-stimulated state), happy, stress, sad, disgust, anger, and surprise. The participants consist of 10 male and 10 female students who watched video programs of two to four minutes in length to stimulate distinct emotions.A new index called the pulse transit time (PTT) has recently emerged and used in research. This signal can be calculated from ECG and PPG signals. We expect to see the accumulate effects of combining ECG, PPG, and PTT features in emotion recognition. Four feature selectors, including fisher discriminant analysis (FDA), sequential backward selection (SBS), sequential forward selection (SFS), and genetic algorithms (GA), were employed to select useful features and reduce feature dimensions, in this study support vector machine was employed as the classifier. Leave-one-out methods were applied to analyze the data recorded from10 male and 10 female participants in cross validation recognizing seven kinds of emotions.Comparing the performance of different combinations of the three categories of physiological signal features, feature selector, and the classifier, we found the combination of the GA feature selector and the SVM classifier achieved the best result. This combination elevated the recognition accuracies in the experimental settings of the leave-one-out cross-validation of 10 males, 10 females, all of the participants and the half-half random selection of all of the participants. All the results demonstrated the combination of the two or three categories of physiological signal features can achieve better recognition rates than that using only one category of physiological signal features. Combination of three categories of features outperforms the combination of two categories of features in emotion recognition, and GA selector plays the major role in promoting the performance.