ses are
a. The
etwork
theses
inected
s done
etwork
' to the
tween
ptimal
her the
iecting
a road
osis in
This
so be
olution
etwork
ie road
have a
tics of
"ments
itrinsic
nto the
ted. In
its two
‚ a new
| these
1serted
topol-
or and
detour
nott.
nc ej
Christian Wiedemann
and the distance along the extracted network (network distance; ) are calculated between all pairs (i, j), # j, of points
which are connected in both networks. The ratio r; ; between these two distances is calculated for each pair (1, j):
network distance’
ref
ij
Tj 2j =
network distance
If r;,; is larger than one, the distance between points P; and Pj along the extracted network is larger than the respective
distance along the reference network. In this case it is referred to as detour factor’, (detour factor with respect to the
reference network). The mean detour factor is defined as the mean of all values detour factor’, ;
The optimum value for the mean detour factor is 1.0. The mean detour factor increases with the amount of missing
connections within connected components of the extracted network and with the degree of “wiggling” extraction (see
Fig. 3).
Reference
Extraction
Figure 3: Wiggling extraction
For the evaluation of the connectivity of the extracted network, a number of points P; are defined equally distributed within
the reference network. All possible pairs of these points are examined if they are connected in the reference network, i.e.,
if they lie within the same connected component. For these CR pairs connected in the reference it is checked whether
they are connected in the extracted network as well. This yields CB pairs which are connected in both networks. Based
on CR and CB, connectivity is defined as
CB # of pairs connected in both networks
connectivity = — = :
> CR # of pairs connected in reference network
The optimum value of the connectivity is 100%. The connectivity decreases with an increasing fragmentation of the
extracted network with respect to the reference network.
The mean detour factor can be used to quantify the improvements achieved by the completion within connected com-
ponents, whereas the connectivity quantifies the improvements achieved by the completion between different connected
components.
5 RESULTS
In this section, results of the approach proposed in this paper are presented. The calculation of the network distance is
performed based on the actual length of the road segments along which the shortest path has been found. No weighting
of the roads according to their class or width is done. The optimal distance is calculated simply as the Euclidean distance
between the respective points, i.e., no additional information like topography, land use or environmental conservation is
taken into account until now. The check of the link hypotheses is performed automatically as described in Sect. 3. In
this step, the geometry of the accepted hypotheses is improved according to the image data. To reduce the amount of
computation time, the search for a road which connects the two end points of a link hypothesis is performed only in a
restricted region of interest (ROI) which contains both end points and which is assumed to contain the connecting road
as well. The following results were obtained using an elliptical ROI with the two end points as focuses and a numerical
eccentricity of 0.75. The iterative process of determining link hypotheses and checking them is broken off automatically
if no unchecked link hypothesis has a detour factor higher than the mean plus three times the standard deviation of the
detour factors of the whole network.
The distance between the equally spaced auxiliary nodes (see Sect. 2.4) was set to 100 m. The maximum optimal distance
was set to 300 m.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 983