Bio-physiological-signals-based VR cybersickness detection

Abstract

With the gradual maturity of virtual reality (VR) technology in recent years, VR industry is in a trend of rapid growth, providing new possibilities for content design. Although VR technology has been able to provide users with excellent immersive experience, side effects that affect the user experience still exist, especially the cybersickness. It would cause extreme physical discomfort to the users and the discontinuation of use. Many researchers have tried to find the inducement of cybersickness and to detect and limit the occurrence of this symptom, but most of the current detection and analysis methods rely on subjective questionnaires to collect users’ posterior states, such as dizziness, nausea, cold sweats, disorientation, eyestrain and so on. There is no mature real-time cybersickness detection system for VR developers to evaluate the susceptibility of their products to cybersickness so far, which has hindered the adoption of VR to some extent. The purpose of this study is to implement the real-time monitoring of cybersickness using physiological sensors to measure data and quantify the influence factors of cybersickness through deep learning model. Besides, we have developed a VR experimental platform and passive navigation task to induce user cybersickness. During the experiment, to train the LSTM Attention neural network model, we collected the user’s real-time physiological signals, including skin electrical activity (EDA) and electrocardiogram (ECG), as well as the position and bone rotation data of the users’ virtual avatar. The model can detect the level of users’ cybersickness in real-time during VR experience. And the model has been verified by the fivefold cross-validation that the average accuracy of 96.85% was achieved for classification of cybersickness level, showing great performance compared with other relevant studies. The results show the feasibility of accurate classification of cybersickness using the model we proposed. Also the model can provide reference for VR researchers and developers to improve the user experience.