A Collaborative Trajectory-Oriented Viewport Prediction for on-Demand and Live 360° VR Video Streaming

Viewport prediction is critical in delivering high-quality 360° virtual reality (VR) videos to a large audience across diverse networks. Traditional computationally expensive viewport prediction mechanisms that rely on static content analysis and individual viewing preferences have proven inadequate in adapting to dynamic content and user preferences. This paper introduces TOPVR, a novel trajectory-oriented viewport prediction approach for both on-demand and live 360° VR video streaming. TOPVR overcomes the challenges introduced by diverse user preferences by leveraging the collaborative behavior of viewers and smartly capturing the relationship between different trajectories. It incorporates a user management system that identifies potential users with similar preferences and the closest trajectory changes to the current user over time. Additionally, a collaborative viewport prediction method estimates future viewing positions for each user based on recent viewing information and similar trajectories and trend changes of other users watching the same content. We evaluate the performance of TOPVR using a real-world head movement dataset recorded using on-demand and live streaming experiments. Our experimental results demonstrate that TOPVR, with its viewer management system and collaborative trajectory prediction mechanisms, outperforms existing benchmark algorithms in terms of higher prediction performance.