'n objects
ind
data
ntext
split
rest,
le.
ntext
1 the
xt object
id object
structure
h for
con-
] the
, the
nent.
'cted
), We
erent
sub-
| of a
tures
jects
Stefan Hinz
will in most cases be possible in high resolution imagery only, because the image features which contribute to the local
context are usually not very prominent. Therefore, the local context is more tightly connected with the high resolution,
whereas information about global context usually can be derived from images with a resolution > 2m and is useful to
guide the road extraction in both scales.
4 EXTRACTION STRATEGY
4.1 Overview
Generally speaking, the extraction strategy inheres knowledge how and when certain parts of the road and context model
are optimally exploited. In some reasonable cases, this knowledge is easy to implement as a set of predefined fixed
rules, e.g., a road never runs through a house (at least in our model). However, the flexibility of rule-based systems is
well-known to be rather limited. Dynamic control systems, e.g., based on Bayesian networks overcome such a drawback.
However, they can get computationally expensive rather quickly. Therefore, we plan to implement a control strategy which
applies dynamic decision making to smaller sub-problems, e.g., finding reasonable parameter settings for the extraction
of markings in shadow regions of a particular scene, whereas the overall flow of the extraction, i.e., (1) Context-based
data analysis, (2) Extraction of salient road segments, and (3) Road network completion is defined as shown in Fig. 4.
Since the road model incorporates 3D information as well as Segmentation of
small sub-structures to a considerable extent, the extraction re-
sisAjeue ejep
peseq-Ixejuon
lies on one side on aerial imagery consisting of overlapping gray Kim Fo TE eens ;
scale images with a fairly high resolution (« 15 cm) and on the udi j| eines 1
other side on an accurate Digital Surface Model (DSM), e.g.,
as it can be derived from dense airborne laser data. In con- Analysis of context relations:
. Road extraction using
trast to other approaches, we neither use orthophotos for extrac- shadow, occlusion,
: : : building outlines, GIS road axes approach forrural areas
tion nor we extract completely in 3D as, e.g., (Gruen and Li,
1997) do, by using multiple images simultaneously. The lat- i
ter procedure is conceptually elegant, but it inherently implies — (.............. Focus on initial Regions of Interest: !
matching procedures throughout the extraction process, which Hypotheses for road axes: ! Hypotheses for road sides: :
1 aci 1 1 x - Bright lines in low resolution images 1 - Fragmented, mosily parallel :
become increasingly burdensome with higher scene complex plait Mei oa ae aM
ities. Furthermore, we want to avoid feature extraction in or- : -Paralle! building outlines E
: 8
thophotos — despite of using accurate DSM information. In- S
accuracies of the DSM due to erroneous height measurements, [............ Constrctionoflanesegments: 2
filtering, resampling, or moving objects still remain and, for in- Extraction of marking groups: ; Detection of vehicle (convoy) outlines: 2
stance, they would disturb useful collinear properties of image NE ro sourate carpe te) g
. . . . , 1 ngular structures o
structures like road markings. Hence, during extraction, we sep- - Parallel marking groups with ' - Analysis of radiometric properties :
arate image and height information to a certain extent: flomogeneous area In between 3 - Analysis of shadow regions ;
At the first stage, down-sampled image information is used > :
to extract the global context regions (see (Baumgartner et al., Fusion based on feno segmante: '
1997)). Contrarily, we use DSM information to analyze height aia ee nn
: : 53 Detection and removal of inconsistencies A
dependent context relations, e.g., for the prediction of shadow — | . ,-.------- piste nants rand
. . . . > s
regions and occlusions. In dense built-up areas approximate | +
road axes and road sides are derived from the DSM, whereas,
. . . . . . Generation of connection hypotheses
e.g. in suburban areas, the image information is used for this. — | [7-777777 a dr rene
. s : Local connections and junctions: 1. Global network connectivity
Thereafter, we project the results of context analysis in each im- S CO E CIM OU coin 2
. . 1 o
age separately. This enables us to treat basic bottom-up proce- - Constraints based on reasonable ! - Detour analysis a
. ‘ I binati ' ©
dures purely as 2D-problems, e.g., the extraction and grouping in g
of linear features for hypothesizing vehicle outlines and for de- ese 2
tecting groups of markings. -- VOTED OL COMET ON IPOS. z
. . . . i : 1 ions: 9
At the next stage, if enough information is collected to construct ORG SANACTON 1098: i ConeMIelfone ¥
. . . . . - Homogeneity tracking + - Shadow :
road objects of a certain spatial extent — in particular, lane seg- snakes E. ie Qoclusin
ments — we include 3D-information by back-projecting the in- - Detection of markings : - Vehicles
dividual objects on the height model. Here, we fuse the results No further hypotheses
of the individual images in order to achieve a consistent data :
set. Optionally, results of road extraction in different context imi”) i
1 . . . Road network
regions can be integrated at this point. Once the lane segments
are fused and aggregated to more complex objects (lanes and fi- Figure 4: Extraction Strategy
nally road segments), connection hypotheses between the road
segments can be generated. Experience from road extraction in the rural context region has shown that this should be done
both locally and globally. Thereafter, verification is carried out in the original images using context relations and stan-
dard verification techniques like snakes and homogeneity tracking. The accepted hypotheses are fused with the already
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 409
