Real-time virtual view rendering for video communication systems using OpenCL

Multi-view video (MVV) offers many possibilities for video communication systems. The creation of virtual views allows to change the viewer's perspective interactively on the receiver side. In this paper we propose a real time algorithm capable of rendering virtual views with a high image quality without the support of scene geometry information. We utilize the highly parallel architecture of recent GPUs for online depth estimation and a sophisticated but generally available shader architecture for efficient rendering. Using these methods we are able to render virtual views with resolutions of up to 3600×3000 with a frame rate of more than 25 frames per second.