ISPRS Commission III, Vol.34, Part 3A ,,Photogrammetric Computer Vision“, Graz, 2002
Figure 4-b. Extracted Roof Regions (Building 2)
4. LIDAR DEM REFINEMENT
For each plane, the contributing points are used to adjust the
plane parameters. The L1 norm (Marshall, 1998) is used in
this step to update the plane parameters. Ll norm
minimization is a robust estimation technique that has the
ability to perform well in the presence of outliers. The
updated parameters serve as the plane parameters at this
stage. The elevations of all points that contribute to a certain
plane are taken as observations, while the unknowns are the
plane parameters.
The next step is to apply the geometric constraints of the
building roofs. The first constraint is the horizontal plane
constraint. The plane parameters are checked to find if any
plane has small slope values, i.e. the values of the parameters
a and b are negligibly different from zero. In this case the
plane is assumed horizontal. The second constraint is the
symmetric constraint. Each two adjacent planes are checked
to find if they satisfy a slope symmetry condition, as one
would expect in a conventional gabled roof. The previous
constraints are used to update the plane parameters. The
mathematical description for horizontal planes is shown in
Equation (2). The complete mathematical description that is
used to find the parameters of two symmetric planes is
shown in Equation (3).
Zyà 7 9X gg 7 P Yga -c=0 2)
a=0 AND h=0
Where n is the number of points in the region.
Zax 7 4; X nd 7 biYya — 6j z0
Z md — A; X m —D ; Ym —Cj =0 (3)
a; +a; =0 OR b+b;=0
Where n is the number of points in region (i), m is the
number of points in region (j). The overdetermined system is
solved by least squares. The adjusted plane parameters are
then used to refine the LIDAR DEM points. Each point
elevation is refined based on the new plane parameters.
Figure 5 and Figure 6 show the LIDAR DEM before and
after the refinement.
A - 105
Figure 6. The Original and Refined DEM for Building 2
5. ROOF BORDERS EXTRACTION
The roof region borders are extracted after finding the roof
regions. Roof border points are points that have at least one
point of its 8-connected neighbors not from the same region
(Rosenfeld and Kak, 1982). Figure 7-a and b show the
extracted roof region borders.