Yuansong Qiao
Athlone Institute of Technology, Ireland
Title: Layer dependency aware multi-view video delivery
Biography
Biography: Yuansong Qiao
Abstract
Multi-view video refers to a composite video stream generated by simultaneous capture from multiple cameras covering different portions, or views, of a scene. The Joint Video Team (JVT) has developed H.264/Multi-view Video Coding (MVC) to enhance the compression efficiency for multi-view video. Streaming of multiview video demands high bandwidth even after encoding. Any loss during transmission will have effect on the real-time quality of experience (QoE) of the end user due to the prediction structure used in H.264/MVC encoder. We will address the challenges in delivering MVC video and introduce MVC delivery technologies in both the traditional client/server based model and peer-to-peer (P2P) based model.
In the traditional client/server based streaming scenario, we have investigated the impacts of network fluctuations (e.g. packet losses) on the quality of streamed MVC video. The test results reveal unexpected differences in video quality amongst the streamed views. An MVC interleaving method is proposed to address this problem, which preferentially transmits the Network Abstraction Layer Unit (NALUs) with higher importance levels for decoding pictures. It reduces transmission errors on more important NALUs and hence enhances the streamed quality of different views.
In the P2P delivery scenario, we have investigated the optimisation problem of maximising outbound bandwidth utilisation of the peers in order to reduce bandwidth usage of the servers. The MVC layer dependency creates challenges in video layer sharing amongst the peers. The layers that can be shared between peers are limited by the layer dependency. A Bit-Torrent based layer-dependency-aware MVC video streaming system has been proposed and evaluated.