n
of
1€
of
Ansgar Brunn
Figure 6: Part of a digital surface model interpolated from laserscanner data with grid width 0.6 x 0.6m? (Copyright by
TopoSys, Ravensburg). The heights are coded in greyvalues.
with
i € {{CP,CP},{CP, EP},{CP,FP}, {EP, EP}, {EP, FP},{FP, FP}},
k € HUH-HM HOY 1H 11/0V37 11.01 107013 and l € (0,1,2,3,4)
e Triangles (faces) (cf. eq. 4):
0.33 053-013 0
Prr (I; |#BL(f)) = (Perl = j|#BL(f) = kW) = | 025 025 025 0935
0 0.66 033: 0
with
j € {H,O,V},k € {0,1,2,3,4}
The likelihood function is defined a priori. We classify introducing the approximate surface and not on the original data
as observations, to reduce the complexity of handling observations in the likelihood function. We choose the following
deduced observations:
e for O-simplices (vertices): the number of normals of touching triangles. For each number of possible amounts of
normals a description-length (Rissanen, 1987) is calculated and transformed into a probability.
e for 1-simplices (edges): the number of normals of touching triangles (cf. 0-simplices).
e for 2-simplices (triangle): the difference angle to the vertical axis.
ir“
a
(a) Approximated building surface with the result of the initial (b) Result of the interpretation and reconstruction. Only
classification coded in intensities. those simplices are shown which are necessary for the ge-
ometric description.
Figure 7: 3D views: Each greyvalue of vertices, edges and triangles represents a classification.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 123