Environmental Engineering Reference
In-Depth Information
3.1
Introduction
the interested readers to find their own way to solve any specific
urban remote sensing problem of their interest.
The use of very-high-spatial-resolution (VHR) satellite imagery
is steadily increasing in many fields from precision mapping
to location-based businesses. The trend toward finer spatial
resolution is pushed by peculiar applications, whose spatial
requirements cannot be met even by high resolution sensors.
Urban-related applications are among the front-runners. How-
ever, it is true that VHR data come with equally numerous limits
and challenges than advantages and improvements. Of course,
this is true for both radar and optical images, but in this chapter
we will focus on the optical data, while in a companion chapter
radar images will be considered.
Indeed, VHR optical imagery is nowadays offered by many
sensors, from Quickbird to Worldview-1 and -2, from Ikonos
II to Geoeye-1, from Cartosat-1 to the EROS (Earth Remote
Observation Satellite) constellation. Moreover, future systems
such as the French Pleiades constellationwill provide faster revisit
times, thus enhancing the timeliness of the data for urban disaster
management and similar applications. A very detailed analysis of
the current situation of VHR sensors for urban applications may
be found in Ehlers (2009) and won't be repeated here. However,
following and updating that source, Table 3.1 summarizes the
main information. It is worth noting that less than 1 m spatial
resolution and less than 1 day revisit images are to be considered
averyclosegoal.
The question we would like to address here is whether the
users (all, and not only ''end-users'') are really able to exploit the
wealth of information coming from these sensors. As researchers
in remote sensing, and especially in urban remote sensing, we
are aware that the answer is generally negative. The reasons are
not only, however, the lack of knowledge by the users of the
potentials of these data, but also the problems and various issues
related to interpretation and (semi)automatic analysis of VHR
imagery. According to what is available in technical literature, we
will attempt in the following sections to address to what extent
these issues are critical, and which is their impact on urban
remote sensing and urban area applications.
To this aim, we will follow this list:
3.2
Geometrical problems
A first group of issues peculiar to VHR images are due to
geometric accuracy. In images whose pixels correspond to a
ground resolution of less than 1 m, a correct georeferencing may
be a real problem. There are a few examples aimed at discussing
how precise the standard products provided by the data providers
are. For instance, inDavis andWang (2001a) many Ikonos scenes
have revealed a consistent RMS (rootmean square) error of nearly
2 m between the referenced and the actual position of the images.
The reasons for these currently challenging precision values
are the reduced availability of same resolution digital elevation
models (DEM) and the approximations in the satellite orbital
parameters. Although the effects of the second problem may be
reduced by accurate post analysis of the satellite position at the
data acquisition time, the lack of accurate DEM is still a problem.
Lidar and aerial systems have been extensively used to obtain
DEMS in urban areas (e.g., Davis and Wang, 2001b; Liu
et al
.,
2007), but their data is often limited to the city centre and do not
cover the extensive areas depicted by VHR optical sensors. The
availability of InSAR (Interferometric SAR) DEMs (Gamba and
Houshmand, 2000) is increasingly being considered as one way
to reduce the problem. Similar results are also obtained using
stereo pair of VHR optical satellite images (Baltsavias, 2005).
A second problem, more peculiar to urban areas, is the need
for orthorectification of VHR images. The fine spatial resolution
in fact causes geometric distortionof the three-dimensional urban
landscape, which must be compensated when using the images
for mapping purposes. There are many examples of approaches
developed for orthorectification of VHR optical images. One
example is Volpe and Rossi (2003), where the problem is put in
the framework of the overall exploitation of Quickbird data.
Finally, the big improvement in new-generationVHR sensors,
which is the capability to steer the sensor, brings further limits
to the usability of these data. In fact, the possibility of different
viewing angles for the same scene is an advantage to reduce revisit
rime (the current 3-day lower limit of many sensors is obtained
due to this capability). It is, however, a further challenge in order
to design analysis approaches that can exploit scenes taken in
different dates with different acquisition geometries and obtain
consistent results. For instance, different viewing angles do not
guarantee the availability of comparable orthorectified products
for the same area. Indeed, occlusions, especially in dense urban
areas, prevent achievingmapping products with similar accuracy.
In turn, this results in erroneous or incomplete change maps,
when these acquisitions are used for change detection after
classification.
Just to provide an example of problems due to slanted viewing
angles, in Fig. 3.1 a Quickbird image covering the downtown of
San Francisco is shown, made available from Digitalglobe after
the 2007 Joint UrbanRemote Sensing Event conference (the com-
plete image set is available at http://tlc.unipv.it/urban-remote-
sensing-2007. In the same figure a very simple classification map
is also provided, and even by a visual inspection it is clear how
the geometric distortion due to the lateral viewing angle with
respect to some of the high-rise building in the area affects the
mapping results.
issues related to geometrical problems of optical VHR data:
all the challenges coming from the geometric accuracy of the
data and its positioning in a common reference system;
●
issues related to spectral problems, especially the lack of
discrimination capability of current VHR sensors and the
need of a compromise between VHR in the spectral and the
spatial sense;
●
issues related to mapping problems, i.e. the need to analyze
the scene no more using a per-pixel approach, but gradually
shifting to a per-object approach;
●
issues related, finally, to multitemporal analysis, e.g. for
change detection, whose validity is strongly connected to
the ability to correlate features and objects more than isolated
pixels.
●
Thefinalpartofthechapterwillprovideanexamplecoming
fromour experience. The approach proposed in those paragraphs
is meant to provide a possible way to overcome some of the
problems discussed in the first sections. Although it is not the
''best'' available methodology, it might be useful to highlight one
or more interesting and possible research paths and thus invite
Search WWH ::
Custom Search