1) get xzy- and rgb- images from kinect (both 3D images) -- this requires already a rough estimate of vert

2) extract horizontal planes by getting peaks in z coordinate

3) reestimte vert from the planes

4) extract all vertical planes by projecting the data onto xy-plane? NO! looses image topology. region growing

-------------

region growing:

compute features for all points
model:
  {set of points} -> params, cost
  new point, params -> delta cost
  del point, params -> delta cost (same, right?)

  
