OPV: BIAS CORRECTION BASED OPTIMAL PROBABILISTIC VIEWPORT-ADAPTIVE STREAMING FOR 360-DEGREE VIDEO

Nowadays the world is more immersive than ever, but huge bitrate and large Internet delay impede the wide applications of 360-degree video. Viewport adaptive streaming is emerging to transmit quality-variable videos based on user's viewport. To prevent the playback stalling, it is necessary to prefetch some subsequent videos within user's future viewport by head motion prediction. However, long-term prediction is easier biased, which results in videos quality drop and quality oscillation. To alleviate the bias' ill effect, it has to decide whether to download a new segment, or replace old tiles with more accurate viewing probability. To this end, we propose a bias correction based optimal probabilistic viewport-adaptive approach, which selects the best tiles and bitrates to alleviate the ill effect of prediction bias. Besides, viewport quality, playback stalling, and quality oscillation are jointly considered in the optimization model to maximize the Quality of Experience in the viewport. The experimental results confirm that our proposal outperforms the existing viewport adaptive approaches.