Motion Sickness Prediction in Stereoscopic Videos using 3D Convolutional Neural Networks

Area of Science:

• Computer Vision
• Neuroscience
• Human-Computer Interaction

Background:

• Virtual reality and 360° videos can induce motion sickness.
• Previous methods for predicting motion sickness relied on motion velocity and depth features.
• A more comprehensive approach is needed to accurately predict motion sickness.

Purpose of the Study:

• To propose a novel 3D convolutional neural network (CNN) method for predicting motion sickness.
• To incorporate user's eye movement as a key feature in motion sickness prediction.
• To improve the accuracy of motion sickness prediction in 360° stereoscopic videos.

Main Methods:

• A 3D CNN model was developed for motion sickness prediction.
• Saliency, optical flow, and disparity maps were used as input features, representing eye movement, velocity, and depth.
• Dataset augmentation techniques, including frame and pixel shifting, were applied to enhance the training data.

Main Results:

• The proposed 3D CNN model achieved more precise predictions of motion sickness.
• The model's predictions showed a stronger correlation with ground-truth sickness data compared to previous methods.
• The inclusion of eye movement features significantly improved prediction accuracy.

Conclusions:

• The developed 3D CNN model effectively predicts motion sickness induced by 360° stereoscopic videos.
• Incorporating eye movement data alongside traditional features enhances prediction accuracy.
• This approach offers a promising tool for mitigating motion sickness in immersive media.