ABSTRACT
The VBR approaches described so far consider each time frame separately, independent of image content at preceding or following time frames. The fundamental physical property of inertia, however, causes all macroscopic objects to evolve continuously over time. By ignoring temporal coherence during geometry reconstruction, no use is made of this regularizing property. What’s more, if temporal coherence is not observed, reconstruction errors at one time step are independent of reconstruction errors at the next time step. The resulting dynamic geometry is temporally inconsistent, which shows up as motion jitter and jumping artifacts during rendering. In contrast, by capitalizing on the fact that motion is continuous, object geometry is recovered more robustly, temporarily occluded regions are plausibly interpolated, and the animated object is rendered coherently.
