Measuring quality of experience for 360-degree videos in virtual reality

Abstract

In recent years, we witness dramatic growing attention in immersive media technologies like 360-degree videos and virtual reality (VR). However, measuring the quality-of-experience (QoE) for 360-degree VR videos is not a trivial task. Streaming such videos to head mounted displays (HMDs) is extremely bandwidth-demanding when compared to traditional 2D videos. In HTTP adaptive streaming, QoE tends to deteriorate significantly during fluctuating network conditions, which results in various bitrate changes and causes multiple stalling events during playback. Thus, understanding how the human visual system perceives 360-degree video with the effect of stalling and different bitrate levels becomes inevitable. In this paper, we investigate the impact of stalling on users QoE under different bitrate levels and the interaction between stalling event and bitrate level for 360-degree videos in VR. To aim this, we first build a 360-degree videos database by encoding videos in three different bitrate levels (1, 5, and 15 Mbps) with 4K resolutions ($3840\times1920$ pixels). We then simulate various stalling events in the videos and conduct a subjective experiment in a virtual reality environment to investigate the human responses. Finally, we use a Bayesian method to estimate and predict the QoE while measuring the quality drop owing to various stalling events and bitrate changes. Proposed solution and prediction results show a strong dependency between playback stalling and bitrate of 360-degree video in VR. Stalling always impacts the QoE of 360-degree videos, but the strength of this negative impact depends on the video bitrate level. The adverse effect of stalling events is more profound when bitrate level approaches to the high and low end, which is in close agreement with subjective opinion.