A GPU-based Framework for Real-time Free Viewpoint Television

Thesis addresses two main problems of Free Viewpoint TV: generation of arbitrary viewpoint in real-time and its delivery to end-user. For the first problem a GPU-based algorithm capable of generating free viewpoints from a network of fixed HD video cameras was developed. We used a space-sweep algorithm to estimate depth information. The view generation sub-system uses it to render new synthetic images for an arbitrary viewpoint. We show how the computations can be divided from each other and parallelized using CUDA. The tests for publicly available Microsoft sequence are provided.;For the second problem, we proposed an open-source browser-based Free Viewpoint Video player. The player is implemented using WebGL where rendering is performed solely on a GPU. It is completely asynchronous - data retrieval from the server and rendering do not block the browser. The player fits perfectly into the presented framework and can greatly improve scalability.