Learning to adapt eye height in 360 videos

Paper Title:

Enhancing Perception and Immersion in Pre-Captured Environments through Learning-Based Eye Height Adaptation

Published on:

24 August 2023

Primary Category:

Computer Vision and Pattern Recognition

Paper Authors:

Qi Feng,

Hubert P. H. Shum,

Shigeo Morishima

Bullets

Key Details

•

Proposes pilot study confirming influence of eye height on perception in 360 video

•

Presents novel deep network to estimate omnidirectional depth and segmentation

•

Uses segmentation to guide layered depth image inpainting

•

Adapts perspective to user's eye height for improved immersion

•

Quantitative, qualitative and user studies validate approach

Explore the topics in this paper

360-degree video

depth estimation

eye height adaptation

image completion

virtual reality

AI generated summary

Learning to adapt eye height in 360 videos

This paper proposes a method to adapt the eye height in pre-captured 360-degree videos to match each user's actual eye level. It uses deep learning on equirectangular and cubemap projections to estimate accurate depth maps. These guide an image completion technique to realistically fill disoccluded areas when changing perspective. Experiments show the approach generates natural views adapted to the user's height.