Compass: A Prefetching Framework with Viewport Patching for 360° Video Streaming

Viewport-dependent streaming strategies are widely adopted in 360 degrees video streaming for bandwidth-saving benefits. These strategies extend a marginal area around the predicted viewport to compensate for inevitable viewport prediction errors. However, existing approaches rely on predefined distribution patterns to extend the marginal area symmetrically, resulting in not only reduced video quality within the viewport but also substantial bandwidth wastage. In this paper, we propose Compass, a prefetching framework with viewport patching for 360 degrees video streaming. Compass simultaneously determines the direction and size of the marginal area based on historical user head movement (HM) velocity and prediction error. Additionally, we introduce a competition-based viewport prediction algorithm and a progressive prefetching mechanism to accurately prefetch video streams before users view them. Simulations conducted using publicly available real user HM trajectories demonstrate the superiority of Compass in improving video quality and saving bandwidth compared to existing schemes.