Viewport-driven rate-distortion optimized scalable live 360° video network multicast

Virtual Reality (VR) technologies enable remote scene 360° video immersion experiences. The growing popularity of VR increases the demand for 360° video delivery over the Internet. Compared to regular videos, 360° videos are characterized by an enormous data volume and spatial user navigation. User interactivity is activated by head movements and changes the spatial portion of a video viewed by the user, hence making only a small portion of the video essential at a time. Therefore, streaming the full video at high quality causes suboptimal use of the bandwidth. The above properties of 360° videos require novel streaming techniques in order to maintain high Quality of Experience (QoE). Live 360° multicast has not being studied yet, due to the emerging nature of 360° video, and it represents one of the most promising applications. Relative to on-demand single user 360° video streaming, it presents new challenges such as handling the interactions of multiple users simultaneously, while dynamically encoding the live 360° video on the fly. We propose a novel scalable multicast live 360° video streaming framework. It comprises a rate-distortion analysis that captures the fidelity-rate trade-offs of 360° videos, an optimization formulation to assign data rates to spatial video regions, and a scalable 360° video data representation. A 2-3 dB of quality gain for lower bandwidth classes and 3-4 dB for higher bandwidth classes are observed over a conventional method that streams the monolithic content at uniform quality. Our work shows that for live 360° multicast achieving quality levels close to on-demand streaming is possible despite the lack of information about future content and user navigation actions.