CRII: NeTS: Data-Driven QoE for Mobile Videos

University Of California-Davis, Davis CA

Investigators

Abstract

Mobile data traffic was around 18 Exabytes in 2013 and is expected to double every year. Of this, 67% is video traffic currently and this percentage is estimated to increase. This necessitates development of strategies to map subjective human scores into on-line deployable methods for Quality of user Experience (QoE) assessment, which can in turn control the underlying service architecture used by network and application providers. The goal of this CISE Research Initiation Initiative project is to develop data-driven mobile video QoE models which can be used for real-time estimation of mobile-video quality in smart devices. The researchers envision collecting large volume of actual video-watch data using a customized application and processing the data to mine important patterns and trends. This work will build a data-driven video QoE model from the user's viewing experience to assess and improve the performance of mobile video applications in contrast to existing work, which has used distortion-specific metrics or full-reference approaches or have data-driven models to model user engagement. The research entails building mobile applications to measure video quality in mobile devices and collect subjective user-experience scores. The anticipated outcomes of this two-year project include answers to key research questions in video delivery for mobile devices: Can we define new objective models for QoE of mobile-multimedia to incorporate the loss in quality caused by freezing, distortions and other factors? Is it possible to develop simpler, more reliable and cost-effective methods for objective evaluation of video quality in battery constraint mobile devices? What is the impact of device aesthetics, network resources and user preferences, and can this impact be accurately modeled? What kinds of guarantees can be made in different application scenarios and what cannot? How can the new metrics be used to design better video delivery system with a focus to enhance user's quality of experience? The researchers will first generate a mobile application to a generate pool of video data obtaining user's subjective video assessment as well as content/network/device/visual quality metadata while allowing a user to play videos from a project-generated dataset or content from popular websites such as Youtube or Netflix. This will be accompanied by development of metrics for quality assessment to capture delivery losses and their proportional impact on video quality. A machine learning model will be developed to model distortions as well as factors such as screen resolution, video resolution, freeze frequency and intensity which impact video quality perception and their proportional impact on video QoE. The development of proposed quality-assessment tools can be used by content and network providers to identify the bottlenecks in ensuring video quality and rectify them with QoE-aware adaptive streaming, caching, transcoding and optimizations. The PI is also committed to including a strong educational plan that has both a conventional component using the project results to drive coursework and involvement of graduate and undergraduates students in research and in building a QoE application for smart phones to involve undergraduate and graduate students.