EEG, ECG, GSR signal based 1D CNN emotion recognition for real time robot interface

초록

Recently, many researches have been conducted on emotion recognition for human-robot Interaction. In order to recognize emotion more accurately, it is necessary to identify inner emotions rather than external ones, and many studies use bio-signals as data. However, in biological signals, especially EEG(Electroencephalogram) collects a lot of data through electrodes over 32 channels. Collecting data from many channels is good for training learning models, but has difficulty in real-time communication between humans and robots. Therefore, in this paper, we reduce the collected EEG data from 32 channels to 10 channels and apply it to the 1D CNN (1 Dimensional Convolutional Neural Networks) learning model along with PNS data. The data used in the experiment was a open database, MAHNOB-HCI, and the raw data of each signal were used as input data for the model. Emotions were categorized as positive, neutral, and negative, and the accuracy of emotion classification through experiments was 62.6%.