User-Driven Encoding and Transmission for Multi-view Video in Successive Motion Model

摘要

Recently, multi-view video that is taken by multiple cameras from different positions and angles to provide the real world experience has attracted more attentions. It allows the users freely change their viewpoints. On the other hand, the traffic of multi-view video is several times larger than traditional multimedia, which brings much more increment in the bandwidth requirement. Compression technologies, such as MPEG and MVC, can greatly decrease the traffic of multi-view video. However, as all the views should be encoded and transmitted, even with MVC bit-rate of multi-view video is still high. In this paper, we analyze users' switch models which are classified into random access model and successive motion model. We find that in the successive motion model only part of frames are possible to be displayed in a period of time. According to the analytical result, we propose a user-driven scheme which can significantly decrease the traffic of multi-view video. In the proposed scheme, only those frames that are possible to be displayed are encoded and transmitted. In order to support this scheme, we also propose a prediction structure as a substitute of the prediction structure in MVC. Evaluation proves that this proposed scheme reduces the average bit-rate for the transmission of multi-view video.