Environmental Engineering Reference
In-Depth Information
non-homogeneity to a wider extent, and even different parts of
the same objects tend to appear as separate scatterers. This adds
to the traditional unreliability of pixel contents in a coherent
radar image, and as a result pixel-by-pixel analysis becomes less
efficient when handling this variety of possible responses within
the same land cover class. Segmentation approaches become a
more reasonable choice (Lombardo
et al
., 2003) and can be
based either on statistical (e.g., Macrı Pellizzeri
et al
., 2003) or
spatial (e.g., Dell'Acqua and Gamba, 2003) analysis. The latter
has a more immediate meaning and is directly related to the
spatial structure of the settlement, which is one of the basic
indicators of its usefulness. It has been shown that formal and
informal settlements differ in the ''order-vs.-disorder'' appear-
ance in remotely sensed images (Niebergall, Loew and Mauser,
2007), and different land use classes can be discriminated even
using coarser data (Gamba, Dell'Acqua and Trianni, 2006), with
a feasible scale-adaptive approach. This latter paper exploits a pri-
ori knowledge about road direction distribution in urban areas:
an adaptive filtering procedure captures the principal directions
of these roads and this information allows enhancing the extrac-
tion results. After road element extraction, a special perceptual
grouping algorithm is devised, exploiting co-linearity as well as
proximity concepts to reduce redundant segments and fill in
gaps. Finally, the road network topology is considered, checking
for road intersections and regularizing the overall patterns using
these focal points.
Another example of urban area delineation can be found in
(Stasolla and Gamba, 2008; Gamba
et al
., 2009) where a map was
extracted from an ASAR frame over the urban area of Beijing
using a combination of morphological and textural methods,
recently developed.
caused by air turbulence, absent in spaceborne platforms, while
on the other hand spaceborne data has to be explicitly corrected
for curvature and rotation of the Earth.
Regardless of the issues connected with a switch in platform,
however, a scaling down of the spatial resolution by more than
one order of magnitude means very different features to emerge,
such as:
●
foreshortening issues appear at a much smaller scale; while
in ERS-like images single buildings were not distinguishable,
in meter-resolution images the different floors of the same
buildings are imaged separately and end up in different
locations in the image;
●
the number of scatterers within the same pixel becomes
smaller and the interference mechanisms change, so do image
statistics;
●
double-bounce phenomena become visible as such, rather
than simply pushing up the average reflectivity sensed on an
urban area.
Some aspects connected to meter-resolution data can be better
explained by referring to representative applications. A selection
of these is presented and discussed in the next subchapters.
5.3.1
Extraction of single buildings
A typical application of high-resolution radar data is three-
dimensional building recognition (Bolter and Leberl, 2000;
Quartulli and Datcu, 2004; Tison and Tupin, 2004; Thiele
et al
.,
2007).
SAR and InSAR data have been widely exploited, for example,
in city cores with high-rise buildings (Gamba, Houshmand and
Saccani, 2000), rural areas, and industrial plants (Simonetto
et al
., 2005). These types of data still show limits, especially when
comparedwiththemoreexpensiveyetmoreaccurateLIDAR
data (Stilla, Soergel and Thoennessen, 2003). LIDAR supplies
3D data with a ground sampling rate of a few points per square
meter and an RMS error on elevation on the order of tenths
of a meter, but it costs between 60-135
¤
per square kilometre.
On the other hand, a spaceborne SAR scene with 8 m resolution
(e.g. RADARSAT-1) is distributed at less than 1
¤
per square
kilometre, but the resolution and height accuracy (on the order
of meters) achievable are definitely lower for spaceborne SAR
when used to produce a DEM in place of LIDAR.
In some cases, those two types of data may be merged to
combinetheprosofboth;purchasingalargeSARscenebutonly
a small parcel of LIDAR data on the same site and comparing
them to derive distortion information which will drive correction
of the rest of the SAR scene may be a good approach to fusion
(Gamba, Dell'Acqua and Houshmand, 2003).
A first reason why SAR is difficult to use is obviously its
side-looking nature, which results in phenomena such as fore-
shortening, layover, occlusion, total reflection, andmulti-bounce
scattering. All these phenomena are also found in natural scenes,
but they show up much more frequently in urban areas. This
happens because the urban environment is rich in smooth pla-
nar patches, often at right angles with each other (resulting in
corner-cube-like reflection) or parallel to each-other (resulting
in several bounces of the incident electromagnetic wave).
5.3
High-resolution SAR
Despite the limited availability of satellite HR SAR data until
recently, techniques that were developed for HR airborne SAR are
also useful for spaceborne HR data. For example: a technique was
initially developed for road extraction from airborne SAR data
(Gamba, Dell'Acqua and Lisini, 2006), combining directional
filtering, perceptual grouping, and topological concepts; this
technique was later applied to spaceborne data with a similar
resolution (Hedman
et al
., 2010). In other cases, the change in
ground resolutionmeans a significant discontinuity in processing
techniques. Features of spaceborne and airborne radar can indeed
be very different.
Although spatial resolution is independent of altitude, viewing
geometry and swath coverage can be greatly affected by altitude
variations. At aircraft operating altitudes, airborne radars must
image over a wide range of incidence angles (up to 70 degrees),
and this has a significant effect on the backscatter from surface
features and on their appearance on an image. Foreshortening,
layover, and shadowing will be subject to wide variations, across
theimage.Spaceborneradarsoperate at much higher altitudes
than airborne radars and are thus able to avoid some of these
imaging geometry problems, with a typical range of incidence
angles from 20 to 50 degrees, also resulting in more uniform
illumination.
More differences are connected with the need for airborne
data to correct artifacts due to the random platform motion
Search WWH ::
Custom Search