Dynamic Viewport-Adaptive Rendering in Distributed Interactive VR Streaming: Optimizing viewport resolution under latency and viewport orientation constraints

Abstract

In streaming Virtual Reality to thin clients one of the main concerns is the massive bandwidth requirement of VR video. Additionally, streaming VR requires a low latency of less than 25ms to avoid cybersickness and provide a high Quality of Experience. Since a user is only viewing a portion of the VR content sphere at a time, researchers have leveraged this to increase the relative quality of the user viewport compared to peripheral areas. This way bandwidth can be saved, since the peripheral areas are streamed at a lower bitrate. In streaming $360\degree$ video this has resulted in the common strategy of tiling a video frame and delivering different quality tiles based on current available bandwidth and the user's viewport location. However, such an approach is not suitable for real-time Interactive VR streaming. Furthermore, streaming only the user's viewport results in the user observing unrendered or very low-quality areas at higher latency values. In order to provide a high viewport quality in Interactive VR, we propose the novel method of Dynamic Viewport-Adaptive Rendering. By rotating the frontal direction of the content sphere with the user gaze, we can dynamically render more or less of the peripheral area and thus increase the proportional resolution of the frontal direction in the video frame. We show that DVAR can successfully compensate for different system RTT values while offering a significantly higher viewport resolution than other implementations. We further discuss how DVAR can be easily extended by other optimization methods and discuss how we can incorporate head movement prediction to allow DVAR to optimally determine the amount of peripheral area to render, thus providing an optimal viewport resolution given the system constraints.