Selected Readings in Vision and Graphics
Dynamic Point Samples as Primitives for
First edition 2004, 180 pages, € 64,00. ISBN 3-89649-949-1
Free-viewpoint video (FVV) uses multiple video streams to re-render a time-varying scene from arbitrary viewpoints allowing for virtual replays and for freeze-and-rotate effects, for instance. In this thesis, we propose the application of dynamic point samples as primitives for FVV by generalizing 2D video pixels towards 3D irregular point samples. The work presented in this thesis was carried out in the context of the blue-c project. blue-c successfully combines the advantages of a Cave™-like projection environment with simultaneous and real-time 3D video capturing and processing of users. The real-time 3D video technology is based on a differential operator scheme which can maintain 3D video by adding, removing, and updating the modified areas of the input video frames only, thereby obviating the need to recalculate the scene’s 3D geometry for each individual frame. The scheme efficiently cuts down the necessary costs for shape computations as well as required network bandwidth. As a result, it allows to transmit a 3D video stream with a mean bit rate between 1.5 and 2.5 Mbps. A novel concept for data acquisition and system control dynamically selects the subset of relevant cameras and adapts to the processing load of the system and features of the target rendering platform.
In addition to the real-time FVV system needed for the blue-c, the recording and efficient coding of geometry-enhanced free-viewpoint video is discussed for off-line FVV. Representations and data formats are presented, which—combined with suitable coding methods—are capable of progressively streaming and displaying sparse multi-view video data from arbitrary viewpoints. The presented image-space framework features multi-resolution, multi-rate and view-dependent decoding. It is based on the fundamental concept of storing all information describing a scene's visual appearance in multi-channel video images. This representation and coding framework is capable of streaming FVV at bit rates as low as 256 kbps and has recently been adopted as future extension of the MPEG-4 AFX standard.
Stephan Würmlin is currently a post-doctoral researcher in the Computer Graphics Laboratory and project leader of the blue-c-II project. He received a diploma degree in Computer Science at ETH Zurich in 2000. From 2000 to 2004, he was a research associate and Ph.D. candidate in the Computer Graphics Laboratory with Prof. Markus Gross. He finished his Ph.D. thesis in May 2004 on the design of the 3D video technology for the blue-c collaborative virtual reality system. His current research interests include free-viewpoint video, point-based representations and rendering, and multimedia coding. The dynamic point sample representation that he developed is adopted by MPEG as an extension of the MPEG-4 AFX standard.
Reihe " Selected Readings in Vision and Graphics " im Hartung-Gorre Verlag
Direkt bestellen bei / to order directly from: Hartung.Gorre@t-online.de