| Summary: | 碩士 === 國立中正大學 === 電機工程研究所 === 102 === This paper proposed an emotion recognition system based on physiological signals. Emotions were recognized by analyzing thirty-second and ten-second electrocardiogram (ECG) signal, photoplethysmography (PPG) signal and pulse transit time(PTT). The system could recognized five emotions including: normal happy stress sad and anger .Ten males and nine females, totally 19 people were included in the tests. They watched four videos, each about 2-4 minutes long to stimulate their emotions. We used three feature selection methods: sequential forward selection(SFS), sequential backward selection(SBS) and genetic algorithms(GA) to select the most effective features to enhance accuracy. Finally, we used support vector machine(SVM) to classify the five emotions. In this paper, both user-independent system and user-dependent system were tested.
In part I, We extracted features from 30 seconds and 10 seconds signal. We found that both signal lengths contributed to similar recognition rate, yet using two physiological signals performed better than using individual ones. The best combination was using features calculated from both ECG and PPG and using GA for feature selection. High recognition rate of 89% in accuracy could be reached by the SVM classifier.
In part II, We tried to apply the same features and GA feature selection method to user dependent system. An average accuracy of 97% recognition rate was achieved. We also tried to reduce feature complexity and number and an average accuracy of 94% could be reached. As high as 100% accuracy could be attained with individual subjects tested with the user-dependent system method.
|
|---|
