International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
done. In order to present a certain level of detail, the tree has to
be traversed down to the given scale level.
AF.c
EF. 0
BC.O.. CEd
AN
A x
/ N
CD. 0 DE 0
Figure 2: Original line (left) and corresponding BLG-tree
(right); the scale levels are indicated in the nodes of
the tree. At the leaves, the scale levels are zero.
For the generation of the BLG-structure, the whole tree has to
be generated in order to give the full zooming from coarse to
fine. The transformation into the SO's is straightforward (confer
Table 1): Starting point is a new line which appears at a certain
scale level EPS, that corresponds to the length of the line sar.
The line is generated by creating point 0 at position A (NPR),
duplicating this point (DV) and moving it to the position F by
increments dx4r. dyar. (MV). At scale level c a new point is
inserted (EPS c). This is accomplished by duplicating point A
(i.e. point 0 in the internal number scheme) and moving it to
position c by increments dxac, dyac, All this information IS
directly coded in the tree. The only issue is an appropriate
sequencing of the insertion of the points, taking the respective
scale levels of the nodes into account.
pes
gure 3: Screenshots visualizing increasing refinement of the
Fig
polygon-visualization (from top left to lower right).
The following Figure 4 shows that the Douglas-Peuker
algorithm is not appropriate for the generalization of structured
objects such as buildings. Therefore, in the next section, an
algorithm for building generalization and the corresponding
decomposition into SO's is presented.
Tim eee
A NN, at 2 t Let ue
Figure 4: Sequence of images of using DP-algorithm to buildin
generalization — which is obviously not suited for
the generalization of such structured objects.
e
e
5.2 Building simplification
Table 2: Coding Douglas-Peuker line simplification.
Figure 3 presents some screenshots of the successive refinement
of polygons using the SO-coding. The iterative refinement is
clearly visible; the user can control the level of detail with the
slider below. Furthermore, the transmission is organized in a
way that only data in the current view will be loaded and
refined.
pov Create new object Building simplification is a special case of a point reduction
EPS $1055) scale level EPS — distance method, where the specific properties of these objects are taken
between points A and F into account. In this case the point reduction is more a structure
NER RA YA Create point 0 with reduction, as properties like parallelism and rectangularity have
coordinates XA and yA to be respected in the algorithm. Here, we used a method that
Dv 0 Duplicate this point -> create analyzes the shape of the building and defines appropriate
point 1 methods to eliminate too small parts of the ground plan, i.e. too
MV 1 (xF-xA) (yF-yÀ) Move point 1 by dx and dy -> short facade elements (see [Sester 2000]). Three different kinds
move it to point F of structures can be identified, for which appropriate reduction
EPS c New event at distance c methods are defined: extrusion or intrusion, offset, and corner
DV 0 Create new point after point 0 (see Figure 5).
by duplicating point 0 S,
MV 1 (xC-xA) (yC-yA) Move this point by dx/dy to Sn+t ouf
point C Ses. edu Sa uu
| Sp+2 | $2 E [23 ]
Sn-2 | | Sn+2 | |
| Sn+1
c)
offset,
Elimination of short facade sn:
intrusion/extrusion and corner.
The decomposition of the changes in terms of SO's is
straightforward. For the example of the offset it is the
following:
An offset consisting of 4 points is replaced by a straight line
consisting of 2 points (see Figure 5a). The reduction process —
which is done when eliminating or generalization this structure
— extends the longest edge adjacent to the short edge s,, in this
case it is line s... A new point is created at the intersection of
the extended line and the predecessors predecessor line (in this
case between line s,,, and line s,;). In order to code this
process in terms of SO's, it has to be inverted, i.e. we start from
the end situation with one line between points | and 4, then
Figure 5:
1296
Internatic
insert po
duplicate
3 new to
EGO for ;
In a simil
two even
Sester, 20
presentati:
(compare
algorithm
larger are:
Figure 6: ]
Fieure 7: T
(
5.3 Typific
Typification
new group \
occur betw
eliminated 2
terms of EG
emerges. Th
5.4 Displac
The coding
very simple,