Impact of 3D Cartesian Positions and Occlusion on Self-Avatar Full-Body Animation in Virtual Reality
(2025) IEEE International Conference on Artificial Intelligence and eXtended and Virtual Reality — Location: Lisbon (27.January.2025)
Files
AIxVR25.pdf
Open Access - Adobe PDF
- 1.13 MB
Details
- Authors
- et. al.
- Abstract
- (en) This study explores methods for enhancing full-body self-avatar animation in virtual reality (VR), particularly by leveraging 3D Cartesian positions of additional body joints. Selfavatar animation is essential to foster embodiment in virtual environments. However, it lacks accuracy when tracking signals are limited to head and hand controllers. We evaluated three state-of-the-art open source Transformer-based models to take advantage of lower- and upper-body joint data and analyze their impact on motion-tracking accuracy. The results show that augmenting the sparse VR inputs with the lower body joints (e.g. feet, knees) significantly improves the accuracy of the avatar, even for upper body motion. However, occlusion of these joints critically degrades the animation quality in every tested configuration, particularly in terms of temporal continuity and smoothness. We show that effectively handling occlusion artifacts is a crucial challenge in this research field.
- Affiliations
Citations
Maiorca, A., Kinart, A., Fletcher, G., Ghasemzadeh, S. A., Ravet, T., De Vleeschouwer, C., Dutoit, T., & et al. (2025). Impact of 3D Cartesian Positions and Occlusion on Self-Avatar Full-Body Animation in Virtual Reality. IEEE International Conference on Artificial Intelligence and eXtended and Virtual Reality, Lisbon. https://hdl.handle.net/2078.5/242167