ABSTRACT
The VBR methods described in the preceding chapter all reconstruct dynamic scene geometry from scratch. In contrast, for many VBR scenarios it is known in advance what kind of object is being recorded. Suitably implemented, such a priori knowledge can be exploited to bias the reconstruction outcome towards plausible results only. To do so, a suitable model of the recorded object must be available. A model also enables the enforcing of low-level as well as highlevel constraints about the object’s motion, from temporally coherent movement to anatomically consistent motion. Model geometry can be highly detailed, which facilitates high-quality rendering results and circumvents rendering inaccuracies due to poorly resolved geometry. In summary, model-based VBR has the potential to attain more robust and accurate rendering results than VBR methods ignorant of recorded scene content.
