A Flexible Viewport-Adaptive Processing Mechanism for Real-Time VR Video Transmission

Recently, with a storm of the concept of virtual reality (VR), panoramic video as the main form of VR is booming. However, one of the major challenges for real-time panoramic video is the high bandwidth consumption, at least 15-20Mbits/s. A truth is that a small portion of the panoramic video which commonly called field-of-view (FOV) can be watched at any time by wearing head-mounted displays (HMD). Therefore, in this paper, in order to address these challenges, we propose a flexible viewport-adaptive transmission mechanism on the server side with a little computational complexity that can ensure the immersive experience of viewers. Firstly, with our method, the FOV area is extracted flexibly from the whole frame at the pixel level. Then, a low resolution frame is obtained by down-sampling from the raw full frame directly. Finally, for the convenience of coding, a series of combination strategies corresponding to the extraction of FOV area are proposed. Experiments results demonstrate that the proposed mechanism can nearly save 55% bandwidth with a little decrease of the quality of viewer's FOV area compared with method of transmitting the whole frame.