MVDN (Multiview Video Delivery over Peer to Peer Networks)MVDN is an extension of the SVDN project to deliver Multiview Video Coding (MVC) video over a P2P network. It is funded by AIT President’s Seed Fund.
Multiview video applications (e.g. 3D television, free viewpoint video application and immersive teleconferencing) are being the next generation of media service instead of traditional 2D televisions. MVC is the state-of-the-art video compression standard for multiview video. MVC encodes multiview video into many view/temporal layers by using intra-view prediction and inter-view prediction. There are some difficulties to deliver MVC video over current networks, e.g. the high bitrate of the encoded multiview video stream even encoded by using MVC and the complicated layer dependencies of the MVC encoded video stream. This project designs a P2P system to deliver MVC video. The system supports view switching.
The objectives of the MVDN system are shown as followings.
Design a P2P system supporting distribution of MVC encoded video.
Enhance the distribution performance by efficiently sharing data between video receivers.
The system is extended from the SVDN project. It is developed based on the Goalbit project and the JMVC reference software. Goalbit is an open source P2PTV platform and JMVC is the software of the MVC codec. The MVC decoder is extracted from the JMVC software and integrated into the Goalbit media player.
The system consists of 4 elements: Broadcaster, Super Node, Peer and Tracker. The broadcaster encapsulates the encoded video content into the newly-designed chunks, which is designed especially for MVC video transmission. Super-peers retrieve the MVC chunks from the broadcaster and distribute to peers. Peers are the end users, which identify the necessary MVC layers, requests chunks from super-peers or peers and decode the encoded video stream for playback. The tracker provides information for peers, e.g. list of peers and access points for view switching. The view switching algorithm is implemented in peers based on the user preferences.
Zhao Liu, Yuansong Qiao, Brian Lee, Enda Fallon