SE-CNN Attention Structure for Quantitative EEG-Based Assessment of VR Motion Sickness.

The applications of virtual reality (VR) technology are currently numerous and promising, but motion sickness (MS) problems are affecting the development of the VR market. Questionnaires are commonly used to subjectively assess motion sickness, but they are applied before and after the user experiences VR and cannot assess the user's motion sickness in real time. In this work, this paper proposes a convolutional neural network (CNN) structure incorporating squeeze and stimulus (SE) attention mechanisms, with subjective questionnaire scores used as markers of electroencephalographic signal (EEG) for real-time prediction of VR motion sickness scores. In this thesis, EEG signals were collected from 16 subjects in a virtual reality environment and the experimental data were fed into the network structure to train the model. The experimental results showed that the root mean square error (RMSE) was 25.69, the Pearson linear correlation coefficient (PLCC) was greater than 0.85, and the Spearman rank correlation coefficient (SROCC) was also greater than 0.85, indicating a significant relationship between the prediction and subjective assessment scores (p<0.05), proving proposed method's effectiveness.