Prediction of Discomfort due to Egomotion in Immersive Videos for Virtual Reality

We consider the problem of automatic assessment of visually induced motion sickness in virtual reality applications. In particular, we study the impact on visual discomfort due to camera motion or egomotion present in the video displayed through a head mounted display. We develop a database of 100 short duration videos with different camera trajectories, speeds and shake levels and conduct a large scale subjective study by collecting more than 4000 human ratings of discomfort levels. The videos are generated synthetically by applying different camera trajectories. We then use the subjective study to learn to predict discomfort by designing features describing the camera motion. The features are based on the ground truth camera trajectory and estimate the camera velocity and shake and depth of the visual scene. We show that these features can be effectively used to predict discomfort by obtaining a high correlation with the subjective discomfort scores provided by humans.