Predictive multicast group management for free viewpoint video streaming

Abstract

Free viewpoint video (FVV) is a new approach of interactive streaming services, where users are able to freely change their viewpoint. The desired viewpoint is synthetized from two or more camera views that must be delivered to the users depending on their continuously changing perspective. Multicast delivery of camera streams is an appropriate solution, however due to network latency and frequent changes of the viewpoint, the required camera streams may arrive too late, interrupting the FVV synthesis and playout. In this paper a seamless FVV streaming scheme is presented based on user viewpoint prediction. In order to avoid the starving of the FVV synthesizer, we propose to use threshold areas to prefetch the camera views that will be probably required for the viewpoint generation. We have formulated the calculation steps of the threshold values in order to minimize the starvation ratio and its duration. The obtained simulation results show that using the predictive multicast management scheme, the clients receive the required camera views in time in more than 95% of the cases. Moreover, we showed that the number of used FVV cameras and the number of clients have significant impact on the performance of the FVV service.