View synthesis framework for real-time navigation of free viewpoint videos

Free viewpoint videos allow users to change the focus of attention arbitrarily. Since it is not feasible to record and compress a large number of videos at all possible angles, most of the views are usually synthesized on demand. In addition, to allow the transition of the view changes smoothly, it is often desirable to render the videos as if only one camera was used from one viewpoint to another continuously. To achieve this feature, a sufficient amount of video frames at various viewpoints between the previously and presently chosen focuses need to be synthesized in a timely fashion. A virtual view can be synthesized using many approaches which lead to different video qualities at the cost of various degrees of computation efforts. In this paper, a synthesis framework for continuous transition between two viewpoints is considered. When the viewpoint is switched from one to another, more viewpoints need to be synthesized for farther distance so that the subjective quality can be smoother.