In this talk I will give an overview of some of our latest
research in the field of 3D scene reconstruction.

The first part of the talk focuses on new performance
capture approaches for high-quality dynamic scene capture.
Our algorithms enable us to jointly reconstruct detailed dynamic scene
geometry as well as dynamic texture information of a moving
person from only a handful of multi-view video recordings.
Our methods succeed without artificial optical markers in the scene and
enable us to reconstruct detailed spatio-temporally coherent
geometry of human subjects in wide everyday apparel, such as a
skirt or a dress. It is also feasible to capture animation models
of animals. We are therefore able to exceed the application range of
previous marker-based and marker-less optical motion capture approaches.

In the second part of the talk I will tap on some of our recent work
on static and dynamic scene capture with active time-of-flight (ToF) 3D
cameras. ToF cameras enable 3D scene capture at video rate and usually also produce good data on scenes where a passive approach, like stereo, would fail. However, current ToF cameras exhibit a non-trivial noise characteristics and only provide very limit X/Y resolution. I will show how detailed 3D models of scenes can be captured despite this at a first glance obstructive characteristics of the sensor.