| Title |
Predicting the Level of Cybersickness via Spatiotemporal Perceptual Features Extracted from Virtual Reality Content |
| DOI |
https://doi.org/10.5573/ieie.2020.57.11.45 |
| Keywords |
Virtual reality; Cybersickness prediction; Spatiotemporal perceptual feature; Supervised learning |
| Abstract |
Although an exploring virtual reality (VR) based on a head-mounted display (HMD) delivers the extended visual experience to a viewer, whose artificially forced stimuli cause inter- and intra- sensory conflicts between visual, vestibular and somatosensory proprioceptor. Such abnormal neurological interaction might provoke a deleterious side-effect, which is called cybersickness. In order to guarantee viewer’s safety, producer and provider have to predict the degree of cybersickness prior to regulating the related attribute of a VR content. Unfortunately, previous researches on cybersickness have only focused on the mutually independent factors. To cope with this problem, we propose a novel scheme for prediction of the level of cybersickness by employing the supervised learning. Towards this, 52 synthetic VR scenes are newly generated, and then the subjective opinions are collected through a large-scale clinical evaluation with participation of 150 volunteers. To capture the robust representations of cybersickness, 17 spatial perceptual features are extracted including the processed raw data which represent the visual-vestibular sensory conflict, the estimated visual sensitivity and complexity from RGB images, and the depth information. Moreover, the extracted spatial perceptual features are transformed to the temporal perceptual features by deploying four pooling methods which reflects the human’s visual characteristics. The extracted features are regressed onto the ground-truth (i.e., the rated score), thus the prediction of quantified level of cybersickness can be achieved by using the trained model, and whose predictive performance has ∼76% correlated relationship with the subjective opinions. |