International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004
3.1.1 Linear Local Context Object In order to fulfill
the requirement of a general framework for describing con-
text objects a generic class Linear Local Context Object
has been defined. All local context objects are defined
based on a common model; they just differ in the relation
to the ATKIS object. The attributive description for any
object defined in Linear Local Context Object is explained
in the following:
e width w(; and certainty width A... The local con-
text objects are represented by means of the center
axis. In order to be able to describe and assess topo-
logic relations knowledge about the width and its cer-
tainty is necessary.
e certainty position projection AÂ,po: In this value the
certainty regarding the position inherited by ortho-
projection is considered: normally orthoimages are
used which have been rectified using a terrain model,
not considering objects above the ground (like trees
and buildings). This leads to a position offset of those
objects in the orthoimage, which has to be taken into
account when topologic relations are assessed. The
precise value of this offset is mostly unknown due to
missing height information, but a range can be speci-
fied.
e certainty position algorithm A540: This value is sim-
ilar to the previous attribute, but it emanates from
the algorithm extracting the object (or more general:
from the source of information): Often it is unclear
how to fix the position of an object. For example the
object Row of Trees is situated beside the road, i.e. the
stems stand outside the carriageway. But in an aerial
imagery one can just observe the crowns and is there-
fore just able to make assumptions about the position
of the stem. This assumption needs to be reflected in
certainty position algorithm.
e precision position and confidence: Refer to the gene-
ral description of the quality attributes. In practice
these measures are obtained for every algorithm indi-
vidually, for example incorporating the pixel size and
sub pixel accuracy for precision position. The confi-
dence is obtained by applying an interior assessment
of the results.
One could also consider defining a Gaussian distribution
for some measures defined as certainty above. If such a
representation suits the requirements better has to be in-
vestigated in the future.
3.1.2 ATKIS Carriageway Object Regarding the AT-
KIS objects the carriageway object is considered. In AT-
KIS the carriageway is implicitly contained in roads and in
objects of higher complexity such as highways. It can be
easily derived from the standardized ATKIS road classes.
The geometric description of carriageways in ATKIS con-
sists of the center axis and the width, given as attribute
(assuming a constant width). The width wA of an ATKIS
object, its certainty A,, 4 as well as the nominal certainty
of the position (A pA)! are given in the attributes of the base
class ATKIS Carriageway Object.
! Normally this value is Ay4 — 3m
804
3.1.3 Extracted Road Object In the relationship model
a general class for extracted road objects is also present: £x-
tracted Road Object. lts attributes are similar to the ones
defined for the Linear Local Context Object, except for
certainty position projection, this certainty is normally not
of interest for roads as they are situated on the ground".
3.2 Relations
In the relationship model the geometric and topologic re-
lations between an ATKIS object and the local context ob-
jects as well as the extracted road objects are also given.
It is important to note that the given relations are indepen-
dent of any quality values: it is a general description of the
reality.
The geometric relation is parallel expresses the fact that
in reality context objects are often parallel to road objects:
For example in open landscapes elongated rows or trees
are situated parallel to roads; in settlement areas the same
holds for building rows.
The topologic relation is important for this work as it must
be taken into account that for example rows of trees must
be situated outside the carriageway given in ATKIS whereas
an extracted road (the surface of the road) must be con-
tained in the ATKIS carriageway and the width of both ob-
jects must be identical. The topologic relations considered
so far are disjoint and contains. The latter one is defined
relative to the ATKIS object. Besides this qualitative topo-
logic relation one may define side conditions. For disjoint
it is often desirable to give a minimum and a maximum
distance (d. 1nin, d. max), which defines on the one hand
an empty space between the road and the respective context
object (d. mn) and on the other side some sort of influence
border (d_max). For example a row of trees must have a
minimum distance to the carriageway (due to security rea-
sons) and also it is expected that trees having a distance to
the carriageway larger than a certain value have no relation
to the road.
The topologic relation contains is for objects being situ-
ated on the carriageway. The possible side condition for
this relation is the indication that the width of both objects
needs to be identical. This is important for extracted road
objects. In Fig. 1 some object classes are derived from Lin-
ear Local Context Object, but an extension to other objects
is possible thanks to the common framework. The given
side conditions for the topologic relations are chosen from
experience, but the incorporation of prior knowledge from
road planning instructions is also possible.
4 STRATEGY FOR ATKIS ROAD ASSESSMENT
AND IMPLEMENTATION
In the framework of road assessment it is sufficient to de-
fine a region of interest (ROI) for each ATKIS Carriageway
Object-Segment including all extracted objects and to as-
sess the given segment using these objects. The size of
the ROI depends on the modeled relations as well as on
the given quality measures: the worse the extracted data,
the larger the ROI. As will be shown later the degree of
support an extracted object gives for the assessment also
depends on its quality. The strategy for road assessment is
as follows:
2Special cases such as road bridges not being present in the height
model are for the moment not of interest.
Internati
I: Ex
sig
pe
te
KI
2 De
jec
Ex
ro
3. As
To
ex
gh
jec
SU]
4. Li
on
ass
The seq
Objects
extracti
input in
al., 200.
reasona
road ex!
procedt
4.4 A
je
The dec
to a cert
extracte
and c) t
an extra
with the
respecti
À is the
Segmen
Ax.
Ayo +
here cot
calculus
area of |
4.0 Rt
The cal
road ob
A A
As the 1
a predef
to the a:
perform
have to
4.3 A:
In the re
tionship
Linear i
Objects.
knowlec
