Viewport-Aware Omnidirectional Video Streaming Using Visual Attention and Dynamic Tiles

PubDate: January 2019

Teams: Trinity College Dublin

Writers: Cagri Ozcinar; Julián Cabrera; Aljosa Smolic

PDF: Viewport-Aware Omnidirectional Video Streaming Using Visual Attention and Dynamic Tiles

Abstract

In this paper, we introduce a new adaptive omnidirectional video (ODV) streaming system that uses visual attention (VA) maps, providing enhanced virtual reality (VR) video experiences. Our proposed method benefits from dynamic tiling and viewport-aware bitrate allocation algorithms. Our main contribution is utilizing the VA maps for deciding the tiling structure (i.e., tile scheme) per chunk and distributing a given bitrate budget to each tile in a viewport-aware way. For this, we first estimate viewport-based VA maps using the collected users’ viewport trajectories. Then, an optimal pair of tiling scheme and unequal bitrate allocation for each tile of a given content is determined per chunk by calculating the expected viewport quality using our proposed VA-weighted objective quality measurement (OmniVA). We evaluate the proposed method performance with varying bandwidth conditions and viewport trajectories from different users. The results show that the proposed method significantly outperforms the existing tiled-based method in terms of viewport-PSNR.