Variational Methods for Stereo Reconstruction

In this talk we will discuss numerous ways that
variational methods may be used to approach the
problem of 3D reconstruction of scenes from multiple
2D images. Jointly estimating both the geometric and
photometric properties of a scene through a common
energy functional yields an algorithm that has so
far performed very nicely on cases where many traditional
algorithms struggle or fail. In addition to reconstruction,
we will also discuss how the same variational framework
can be applied to refining camera calibration parameters
and even color calibration parameters to make different
images of a common scene more consistent with each other.
Along the way, we will point out some interesting mathematical
observations and findings about when the joint evolution of
a surface along with scalar function supported on the evolving
surface is ill posed.