Gaze-Based Personalized Multi-View Experiences

Abstract

This paper describes a solution for delivering andpresenting stereoscopic video content to users in aninnovative way. It adopts the multi-view paradigm of theH.264-MVC video coding standard and the emergentMPEG DASH specification to provide users inheterogeneous network environments multiple and varyingperspectives of stereoscopic video sequences. Unlike existing3D systems based on multi-view technology, which requirehigh transmission bandwidth and high processing power onthe terminal device to achieve the same objective, theproposed solution is able to make an efficient use of networkresources whilst being cost-effective. It offers users a higherquality of experience by seamlessly adapting the quality ofthe delivered video content according to the networkconditions, whilst providing a more realistic sense ofimmersion by offering stereoscopic views of the scene,dynamically switching the perspective to match the interestsof the user. A non-intrusive head-tracking system using anoff-the-shelf Web camera detects the focus of attention ofthe user, transmitting this information to the server thatselects the most appropriate view to send to the client.Additionally, the system is able to generate the multipleperspective stereoscopic scenes using 2D cameras.