Predicting Stalling Events in Streaming Videos

Given the tremendous increases in video traffic, which account for the majority of mobile data traffic, there has been a dramatic shift towards over-the-top video streaming. Due to limits on wireless network capacity, and with an increasingly knowledgeable base of consumer users demanding higher quality video display services, accounting for an end user's quality of experience (QoE) has become an essential measure of network performance. QoE refers to a viewer's holistic perception and satisfaction with a given communication network service.

WNCG students Deepti Ghadiyaram and Janice Pan and WNCG professor Al Bovik have developed a model for predicting continuous-time QoE for videos that may contain interruptions in video playback, also referred to as rebuffering or stalling events. Their model is dynamic and perceptually-based, accounting for the hysteresis effect caused by a stall, which is the hypothesis that a viewer's recent levels of satisfaction also impact their instantaneous QoE. A US Patent has been applied for.

This model is the first to consider the inherent memory or ‘recency effect’ in the human visual system while also accounting for lengths, frequency of occurrence, and positions of stall events - factors that interact in a complex way to affect a user's QoE. Developing models that can accurately predict an end user's QoE can enable the more efficient design of quality-control protocols for video streaming networks to reduce network operational costs while still delivering high-quality video content to customers.