GRATED
A
rban areas is a
n the difficulty
th this is to use
rtainty can be
Ight Detection
t model of the
S (i.e., parking
id existing are
followed by an
rmation of the
corresponding
town area. The
ccurate by the
cessing of the
mating road
| problems of
licated to be
lied to extract
0 some extent
nodel. For the
lly from high
which lead to
cene and road
buildings and
information,
the principle
gh reliability
is a relatively
to traditional
lenty of scene
h as roads and
us reflectivity
re surface in
he analysis of
as (Hofmann,
urban areas be
001; Alharthy
t information
to reason if a
the missing
n cases when
r shape can be
intensity data
ser is suitable
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
for ground materials. The relative separations between ground
features (i.c., asphalt road, grass, building and tree) have been
compared using intensity data. It is found that the separabilities
are very high for road vs. grass and road vs. tree (Song ct al.,
2002). In many cities, road networks are arranged in a grid
structure in urban areas. These grid roads are mainly composed
of parallel and orthogonal straight roads with respect to the
main orientation of the network. The existence of streets can be
detected much more easily from the arrangements than from
imagery in which the highly complicated image content and
lack of information lead to high complexity of direct extraction
of the street network. It is recognized that the simple geometry
and topology relations among grid streets may be used to
improve the reliability of road extraction results significantly.
As mentioned above, instead of using imagery, using lidar data
can be easier to extract the road primitives in built-up areas,
while imagery can also be used for additional information for
verification and accurate extraction. Many clues of road
existence can be obtained from high resolution imagery. The
motivation of this paper is to explore the strategy and
methodology of integrated processing of lidar data and high
resolution imagery in order to obtain reliable road network
information from the dense urban environment. In the followed
section, the case study data is introduced and the overall
strategy of the processing is given. The third section describes
the road extraction methods by using of these two source of
information, including road area segmentation, road clue
detection and verification, fusion of the clues from the two data
sources. The case study result is presented and conclusion
remarks are then given.
2. OVERVIEW OF INTEGRATED PROCESSING OF
LIDAR AND HIGH RESOLUTION IMAGERY FOR
ROAD EXTRACTION
2.1 Data of the Case Study Area
In early 2002, Optech International, Toronto completed a flight
mission of acquiring the lidar data of Toronto urban area using
its ATLM 3200. The lidar dataset provided is around downtown
region. The roads in the study area are coated with asphalt with
pebbles or concrete. The first and last returns lidar range and
intensity data were collected. The dataset contains about 10.6
million points and has a density of about 1.1 points/m“. We
generate the DTM using the last-return lidar range data, and
also obtain the height data by subtracting the DTM from the
range data (Hu, 2003). The height data contains height
information that has removed the retain relief relative to the
bare Earth, and puts all the ground features on a flat reference
plane. Figure 1 (a) and (b) shows the first-return intensity data
and the height data. The high resolution imagery is obtained
from ortho-rectified aerial image of the same area. The image
resolution is 0.5m. To do integrating processing, it is re-
sampled into Im resolution and is manually registered with the
lidar data in geometry. Figure 1 illustrated the lidar data of the
area. Figure 1 (¢) shows an image window of the high resolution
imagery. Its size is 1024 by 1024 pixel. Considering the
computational cost, we carry out our extraction in this selected
area, which demonstrates typical scene of dense urban area. It
contains buildings with great height, roads (streets) and many
kinds of typical ground objects (parking lots, grass land, trees,
vehicles etc.). It is feasible to testify our method.
(c) High resolution aerial imagery
Figure 1. Lidar data and imagery used for road extraction
2.2 PROCESSING WORKFLOW
Figure 2 illustrated the workflow of the integrated processing
for road network extraction from the dense urban environment.
The strategy is based on an observation to the scene in which
the major clues of road existence should be from lidar data from
which the height data enables it eliminate the principle
difficulties in occlusion of the roads. So firstly the lidar data are
used to obtain the candidate road stripes, because in the build-
up area the dense building arrangement demonstrates grid
structure and the grid road network can be perceived easily from
the structure rather than from optical imagery due to the
occlusion. The segmented road and open areas could be further
segmented by using of the results of classification of the optical
imagery. The grass lands and tree areas are extracted from the
image by pixel based classification. The open areas extracted
from the lidar data contain road stripes and parking lot areas.
Possible parking areas are extracted by morphologic operation
of the segmented lidar data. To verified and differentiate the
road stripes and the parking areas, clues from shape analysis
and vehicle detection are involved. The vehicle detection is
fulfilled in the high resolution optical imagery. The information
of verified roads and parking areas is used for formation of the
road grid.
In the next section, the methods of integrated processing are
briefly described.
