Tile-Based Qoe-Driven Http/2 Streaming System For 360 Video

Abstract

Recently, the 360-degree video has become a hot topic in multimedia area. However, the requirements of high bitrate, low Internet interactive latency and high perceived quality limit its further applications. So in this paper, we design a tile- based QoE-driven HTTP/2 streaming system for 360 video. It uses a new projection method, Sinusoidal Projection (SP), to reduce the bitrate of tiles. A novel cross-user's behavior learning method is also used to predict viewpoint. To improve bandwidth utilization, it pushes multiple tiles in one request by HTTP/2. Besides, by using a QoE-driven framework, our approach can significantly improve users' perceived quality. The numerous experiment results have demonstrated the efficiency of the proposed system. Compared with the legacy methods, the transmission bitrate drops about 17%, the view- port prediction accuracy improves 30%, the Viewport-PSNR improve 22% and the transmission latency drops about 30%.