From markerless motion capture to unstructured video based rendering
Luca Ballan
ETH Zuerich
Talk
Luca Ballan is a research scientist at the Institute of Visual Computing at ETH Zurich. In 2005 he finished his computer engineering studies at the University of Padova, Italy and in 2008 he received his PhD degree from the same university with a dissertation on marker-less motion capture. Luca's research interests include passive 3D reconstruction of static and dynamic scenes, in particular estimating shape and pose of humans from images or
videos, and video based rendering.
The natural approach for free-viewpoint video (FVV)
of a performance is to model the geometry and appearance
of the entire scene. While this is possible in indoor scenarios
by controlling all the parameters, from lighting to background,
only weak assumptions can be made for outdoor performances
captured by for example members of the public.
This talk is structured in two parts. First I will talk about my
past work on high-accuracy marker-less motion capture in
indoor scenarios. In the second part, I'll present our new
work on unstructured video based rendering (VBR) that
explores several simplifying assumptions in order to make
limited video navigation of extemporaneous performances
more tractable. We demonstrate that performances filmed
using hand-held multi-view cameras can be "browsed"
to some extent by optimizing when to perform view interpolation.