3D reconstruction and camera pose estimation (structure from motion)
from perspective video is nowadays a well established technique.
However, there are still some problems and ambiguities associated with
this technique. In case of perspective cameras with small field of view,
ambiguities between translation and rotation estimation occurs due to
data correlation. This problem might be tackled with wide fov fisheye
lenses. In addition, to reliably estimate structure and pose, sufficient
camera motion and feature tracking over long sequences is needed. Some
of the problems associated with sfm can be solved by incorporating an
active 3D range camera that delivers 3D depth per pixel at realtime
frame rates. This will allow to obtain absolute 3D scene structure and
pose without constraints on the camera motion and field of view. In my
presentation I will talk about the combination of a 3D range camera with
one or multiple 2D cameras (small fov perspective and wide fov fisheye),
how to calibrate such systems, and show applications in 3D pose tracking
and reconstruction.