Optimizing Quality Of Experience Of Free-Viewpoint Video Streaming With Markov Decision Process

Free-viewpoint video (FVV) streaming is an emerging type of multimedia service that provides unprecedented interactivity for users to explore different scenes at different angles. Compared with traditional 2D video, delivering FVV is much more challenging because the uncertainty of view switching together with bandwidth variation may cause frequent playback interruptions. In this paper, we present the first stochastic optimization on the quality of experience (QoE) of FVV streaming that consists of two key components: video quality and interruption time. The bandwidth dynamics and the view switching are captured by different finite-state Markov processes, respectively. We formulate a novel Markov decision process (MDP) model where the state as well as its transition incorporates the view and the utility considers the quality of watching other than downloaded chunks. A policy iteration approach is adopted to obtain the optimal chunk selection strategy. Extensive simulations manifest that our approach outperforms the bandwidth based counterpart in terms of average bitrate and number of interruptions.