fusion of optical and radar remote sensing data: munich city example

fusion of optical and radar remote sensing data: munich city example
In: Wagner W., Székely, B. (eds.): ISPRS TC VII Symposium – 100 Years ISPRS, Vienna, Austria, July 5–7, 2010, IAPRS, Vol. XXXVIII, Part 7A
Contents
Author Index
Keyword Index
FUSION OF OPTICAL AND RADAR REMOTE SENSING DATA:
MUNICH CITY EXAMPLE
G. Palubinskas *, P. Reinartz
German Aerospace Center DLR, 82234 Wessling, Germany - (gintautas.palubinskas, peter.reinartz)@dlr.de
KEY WORDS: Fusion, Imagery, Multisensor, SAR, Optical, Orthoimage, Acquisition, Geometry
ABSTRACT:
Fusion of optical and radar remote sensing data is becoming an actual topic recently in various application areas though
the results are not always satisfactory. In this paper we analyze some disturbing aspects of fusing orthoimages from
sensors having different acquisition geometries. These aspects are errors in DEM used for image orthorectification and
existence of 3D objects in the scene. We analyze how these effects influence a ground displacement in orthoimages
produced from optical and radar data. Further, we propose a sensor formation with acquisition geometry parameters
which allows to minimize or compensate for ground displacements in different orthoimages due the above mentioned
effects and to produce good prerequisites for the following fusion for specific application areas e.g. matching, filling
data gaps, classification etc. To demonstrate the potential of the proposed approach two pairs of optical-radar data were
acquired over the urban area – Munich city, Germany. The first collection of WorldView-1 and TerraSAR-X data
followed the proposed recommendations for acquisition geometry parameters, whereas the second collection of
IKONOS and TerraSAR-X data was acquired with accidental parameters. The experiment fully confirmed our ideas.
Moreover, it opens new possibilities for optical and radar image fusion.
to optimize it if necessary. Of course having not a full control
on sensors as in a military community it is not so easy but is
still possible to influence some acquisition parameters. In this
paper we analyze the effect of ground displacements in
orthoimages of optical and radar sensors due to the height error
in the DEM used during orthorectification process and 3D
objects characteristics (height) for various data acquisition
parameters such as sensor look angle (elevation) and look
direction, satellite flight direction and sun illumination
direction.
The paper is organized in the following way. First, the
methodology used for the proposed approach is presented in
detail. Then, data used in experiments are described, followed
by the presentation of experimental results, conclusion,
acknowledgments and finally references.
1. INTRODUCTION
Data fusion is an extremely emerging topic in various
application areas during the last decades. Image fusion in
remote sensing is one of them. However fusion of different
sensor data such as optical and radar imagery is still a
challenge. In this paper the term ‘radar’ is equivalent to
Synthetic Aperture Radar (SAR). Though the data fusion is well
spread over different communities there are quite few attempts
of its definition. The first one is the so called JDL information
fusion definition (U.S., 1991) popular in military community.
This definition is based on the functional model including
processing levels and full control on sensors thus making it
difficult to transfer to other communities. Another data fusion
definition more suitable for a broader community is introduced
in (Pohl, 1998) mainly emphasizing (and thus simultaneously
limiting to) methods, tools and algorithms used. A more general
definition is proposed in (Wald, 1999; Data Fusion Server) as a
formal framework in which are expressed the means and tools
for the alliance of data originating from different sources.
According this definition an alignment of information
originating from different sources now becomes a part of the
fusion process itself.
There exist numerous remote sensing applications e.g. image
matching and co-registration (Suri, 2008), pan sharpening
(Klonus, 2008), orthoimage generation, digital elevation model
(DEM) generation, filling data gaps, object detection,
recognition (Soergel, 2008), reconstruction (Wegner, 2009) and
classification (Palubinskas, 2008), change detection, etc which
are already profiting or can profit significantly from a data
fusion.
For the fusion of data from sensors exhibiting different
acquisition geometries such as optical and radar missions it is
important to understand their influence on a fusion process and
2. METHOD
In this section we analyze two effects: height error in DEM
used during orthorectification process and 3D object height and
their influence on ground displacements in orthoimages from
optical and radar sensors. The study results in a proposal of
several data acquisition parameters: sensor look angle
(elevation) and look direction, satellite flight direction and sun
illumination direction leading to an optimal sensor formation
for the following optical and radar data fusion.
2.1 Effect of DEM height
Ground displacement Δx due the height error Δh in the DEM
for an optical and a radar sensor orthoimage is shown in Figure
1.
* Corresponding author.
181
In: Wagner W., Székely, B. (eds.): ISPRS TC VII Symposium – 100 Years ISPRS, Vienna, Austria, July 5–7, 2010, IAPRS, Vol. XXXVIII, Part 7A
Contents
Author Index
Keyword Index
Optical/Radar Sensor
Optical/Radar Sensor
θ
θ
Δh
Δh
Δxopt
Δxrad
Δxrad
Δxopt
Figure 2. Ground displacement Δx for a 3D object of Δh height
for an optical and radar sensor orthoimage. Look directions:
pink line for optical sensor, blue line – radar sensor. The green
horizontal line stands for a flat DEM, which doesn’t include
height information of objects. The green circle stands for a true
ground position of a 3D point, whereas the red circle – a
displaced position. Thin black line perpendicular to blue line
shows approximately the radar wave propagation. Flight track is
into plane.
Figure 1. Ground displacement Δx due the height error Δh
(positive and negative) in a flat DEM for an optical and radar
sensor orthoimage. Look directions: pink line for an optical
sensor, blue line – radar sensor. The green horizontal line stands
for a true DEM, whereas the red line stands for an error in the
DEM (same for both sensors). Similarly, the green circle stands
for a true ground position of a 2D point, whereas the red circle
– a displaced position. Thin black lines perpendicular to blue
line show approximately the radar wave propagation. Flight
track is into plane.
2.3 Equality of displacements
Ground displacements are equal to
Δxopt = Δh ⋅ tan θ opt
We have seen in the previous sub-sections that sizes of ground
displacement are different (different formulae) for optical and
radar sensors and, moreover, displacement directions are
opposite for different sensors. The size equality of ground
displacements
(1)
for optical sensors and
Δxrad =
Δh
tan θ rad
Δxopt = Δxrad
(2)
(3)
is fulfilled for the following sensor look (elevation) angles
for radar sensors. We have to note, that ground displacements
are towards the sensor for the optical case and opposite for the
radar case (sign of displacement is ignored in formulae). For
details on radar geometry see e.g. (Oliver, 1998).
θ opt + θ rad = 90°
(4)
We have to note, that smaller ground displacements are
obtained in case of
2.2 Effect of 3D object height
θ opt < θ rad
Ground displacement Δx for a 3D object of Δh height for an
optical and a radar sensor orthoimage is shown in Figure 2.
(5)
In order to compensate opposite displacement directions for
different sensors the look directions of different sensors should
be opposite. Under the conditions of (4) or (5) structures in
optical and radar images appear almost in the same positions
thus leading to an easier interpretation and further processing of
joint data.
Formulae for ground displacements are the same as in the
previous sub-section: for optical case equation (1) and radar
case - (2). The only difference is a displacement direction: it is
away from sensor for the optical case and opposite for the radar
case.
182
In: Wagner W., Székely, B. (eds.): ISPRS TC VII Symposium – 100 Years ISPRS, Vienna, Austria, July 5–7, 2010, IAPRS, Vol. XXXVIII, Part 7A
Contents
Author Index
Keyword Index
TS-X see (Eineder, 2005). Other scenes of the same urban area
of TerraSAR-X and IKONOS have been ordered from existing
archives.
2.4 Optimal sensor constellation
In this sub-section we propose an optimal optical and radar
sensor formation for an image acquisition compensating/
minimizing ground displacement effects of different sensors
(see Figures 3, 4). A sum of look angles should give
approximately 90° (Figure 3).
Radar sensor
Radar sensor
Optical sensor
Flight direction
Flight direction
N
Optical sensor
Earth
Look direction
θrad
θopt
S
θopt+θsar=90°
Sun
Figure 4. Proposed optical and radar sensor formation is
illustrated. Flight directions should be parallel, in same
direction and perpendicular to look directions which are
opposite for different sensors (right drawing). Sun illumination
direction is from an optical sensor to the target on the Earth.
Earth surface
Figure 3. Proposed optical and radar sensor formation is
illustrated. A sum of look angles should give 90°.
4. EXPERIMENTS
Flight directions should be as parallel as possible and
perpendicular to look directions which are opposite for different
sensors (Figure 4). Same flight directions are not required in
general e.g. airborne case. A sun illumination direction is from
an optical sensor to the target on the Earth in order to see a side
of a 3D object which is in shadow in radar image and thus
enable full reconstruction of a 3D object. This sensor
configuration allows a recovery of 3D object shadows during
further data fusion, except a case when the Sun illumination
direction is the same as for SAR look direction. Displayed left
looking radar and right looking optical sensor formation can be
preferable due to the Sun illumination direction which is from
an optical sensor to the target on the Earth in order to see that
side of a 3D object which is in shadow in the radar image and
thus enable full reconstruction of a 3D object. Of course, the
second sensor formation with a right looking radar and left
looking optical sensor can be useful for data fusion too.
Our approach could be applied in both airborne and space
remote sensing. As an example we consider the latter one.
Currently, most space optical remote sensing satellites are
acquiring data in descending mode, so a radar satellite should
also acquire in a descending orbit. Thus both satellites would
fly in the same direction (quasi-parallel orbits). The
requirement of opposite look angles and a special sun
illumination direction result in a left looking radar sensor and a
right looking optical sensor what is achievable with current
radar missions though not in a nominal mode (left looking
radar). Additionally, larger look angle of SAR sensor than look
angle of optical sensor allows minimizing the sizes of ground
displacements.
Two experiments, one with a proposed sensor formation and
one with an accidental sensor formation were performed to
show the potential of our approach. The optical image has been
corrected for absolute position by ground control, which
yielded a global shift value of approximately 10 m in xdirection for the WV-1 data and 6 m in x-direction and 2 m in
y-direction for the IKONOS data in comparison to image
rectification without ground control. TS-X data Enhanced
Ellipsoid Corrected (EEC) product can be used without ground
control, since absolute positioning Root Mean Square Error
(RMSE) for the Spotlight mode is in the order of 1 m
(Bresnahan, 2009).
4.1 Proposed sensor formation
Scene parameters for the proposed sensor formation experiment
are presented in Table 5.
Sensor
Parameter
Image data
Image time (UTC)
Mode
Look angle
Polarization
Product
Resolution gr x az (m)
TS-X
WV-1
7-Jun-2008
05:17:48
Spotlight HS
49.45° Right
VV
EEC
1.0 x 1.14
18-Aug-2009
10:50:42
PAN
38.3° Left
L2A
0.89 x 0.65
Table 5. Scene parameters of the first experiment over Munich
city
3. DATA
Part of Munich city acquired by WV-1 (upper image) and TS-X
(lower image) using the proposed satellite formation is shown
in Figure 7. Yellow grid lines are for better orientation between
the two images. Ground objects like streets and plazas as well
as structures e.g. buildings and trees can be easily detected in
both images and are found at the same geometrical position in
both images. Only the feet of the buildings, which are
differently projected in the radar image due to foreshortening
The German Aerospace Center DLR and DigitalGlobe have
been engaged in a modest R&D project to investigate
complementary uses of Optical and Radar data. Coordinated
collections of high resolution TerraSAR-X (TS-X) and
WorldView-1 (WV-1) data during July-August 2009 have been
acquired. For this experiment one scene of WorldView-1 over
Munich city, Germany has been acquired. For more detail on
183
In: Wagner W., Székely, B. (eds.): ISPRS TC VII Symposium – 100 Years ISPRS, Vienna, Austria, July 5–7, 2010, IAPRS, Vol. XXXVIII, Part 7A
Contents
Author Index
Keyword Index
are found at different positions. The roofs and tree crowns are
well in place and can be overlayed correctly for any further
processing. Groups consisting of 2, 5 and 6 buildings are
highlighted in blue color in both images to show a good
correspondence.
REFERENCES
Bresnahan, P., 2009. Absolute Geolocation Accuracy
Evaluation of TerraSAR-X Spotlight and Stripmap Imagery –
Study Results. In: Proceedings of Civil Commercial Imagery
Evaluation Workshop, 31 March – 2 April 2009, USGS, Fairfax
Virginia, USA.
4.2 Accidental sensor formation
Scene parameters for the accidental
experiment are presented in Table 6.
Sensor
Parameter
Image data
Image time (UTC)
Mode
Look angle
Polarization
Product
Resolution gr x az (m)
sensor
Eineder, M., Schättler, B., Breit, H., Fritz, T. and Roth, A.,
2005. TerraSAR-X SAR products and processing algorithms.
In: Proc. of IEEE International Geoscience and Remote Sensing
Symposium (IGARSS’05), 25-29 July, 2005, Seoul, Korea,
IEEE, vol. VII, pp. 4870-4873.
formation
TS-X
IKONOS
25-Feb-2008
16:51:15
Spotlight HS
22.75° Right
VV
EEC
1.6 x 1.3
15-Jul-2005
10:28:06
PAN
5.0° Right
Orthoimage
0.8 x 0.8
Klonus, S., 2008. Comparison of Pansharpening Algorithms for
Combining RADAR and Multispectral Data. In: The
International Archives of the Photogrammetry, Remote Sensing
and Spatial Information Sciences. Volume XXXVII, Part B6b,
Beijing, China, pp. 189-194.
Palubinskas, G. and Datcu, M., 2008. Information fusion
approach for the data classification: an example for ERS-1/2
InSAR data. International Journal of Remote Sensing, vol.
29(16), pp. 4689-4703.
Table 6. Scene parameters of the second experiment over the
city of Munich
Again, part of Munich city acquired by IKONOS (upper image)
and TS-X (lower image) using the accidental satellite formation
is shown in Figure 8. Yellow grid lines are for better orientation
between two images. For this case it is quite difficult to find
corresponding structures in the two images. Only ground
objects like streets can be found at similar places but buildings
are represented in very different geometry and can be hardly
allocated to each other. Also from a radiometric point of view
the differences are higher than in Figure 7 probably due to
different shadow properties. The same groups consisting of 2, 5
and 6 buildings as in sub-section 4.1 are highlighted in blue
color in both images again. In this case it is quite difficult to
identify the same number of buildings in both images.
Pohl, C. and van Genderen, J. L., 1998. Multisensor image
fusion in remote sensing: concepts, methods and applications.
International Journal of Remote Sensing, 19(5), pp. 823-854.
Oliver, C. and Quegan, S., 1998. Understanding Synthetic
Aperture Radar Images. Artech House, Boston.
Suri, S. and Reinartz, P., 2008. Application of Generalized
Partial Volume Estimation for Mutual Information based
Registration of High Resolution SAR and Optical Imagery. In:
Proc. of 11th International Conference on Information Fusion
(FUSION’2008), June 30 – July 3, 2008, Cologne, Germany,
pp. 1257-1264.
The Data Fusion Server. Available from: http://www.datafusion.org/ (Accessed 24 February 2010)
5. CONCLUSIONS
U.S. Department of Defense, Data Fusion Subpanel of the Joint
Directors of Laboratories (JDL), Technical Panel for C3, "Data
fusion lexicon," 1991.
In this paper we address a problem of fusion of optical and
radar remote sensing imagery. Alignment of information
coming from different sources is an important prerequisite for
the following fusion in various applications. Especially for a
rapid fusion of optical and radar data a specific imaging is of
advantage. We propose an optical and radar sensor formation
which accounts for different acquisition geometries and
minimizes displacements for ground and 3D-objects in
orthoimages of optical and radar sensors. The preferred sensor
formation is a perpendicular viewing from the two sensor
systems due to the complimentary nature of their viewing
geometries. For this case the image geometries are nearly
independent to errors in the underlying DEM and especially to
buildings or other 3D objects, not represented in the DEM. A
fast and consistent overlay of the two data sets for on ground
and other surfaces is reached. As an example two pairs of high
resolution optical (WorldView-1 and IKONOS) and radar
(TerraSAR-X) images have been acquired over the urban area Munich city in Germany – for different sensor formations.
Results show a great potential of the proposed approach for
further applications of data fusion with optical and radar
instrumentation since the geometric positions of the objects can
be observed at the same absolute position.
Wald, L., 1999. Some terms of reference in data fusion. IEEE
Transactions on Geosciences and Remote Sensing, 37, 3, pp.
1190-1193.
Wegner, J.D., Auer, S., and Soergel, U., 2009. Accuracy
Assessment of Building Height Estimation from a High
Resolution Optical Image Combined with a Simulated SAR
Image. In: Proc. of ISPRS Hannover Workshop 2009 - High
Resolution Earth Imaging for Geospatial Information, 2-5 June,
2009, Hannover, Germany, ISPRS, vol. XXXVIII-1-4-7, part
W5.
ACKNOWLEDGEMENTS
We would like to thank DigitalGlobe for the collection and
provision of WorldView-1 scene over Munich city,and
European Space Imaging for providing the IKONOS-2 scene.
184
In: Wagner W., Székely, B. (eds.): ISPRS TC VII Symposium – 100 Years ISPRS, Vienna, Austria, July 5–7, 2010, IAPRS, Vol. XXXVIII, Part 7A
Contents
Author Index
Keyword Index
rg
az
rg
az
Figure 7. Part of Munich city acquired by VW-1 (upper image) and TS-X (lower image) using the proposed satellite formation.
Yellow grid lines are for better orientation between two images. Red arrows show flight (az) and look (rg) directions.
185
In: Wagner W., Székely, B. (eds.): ISPRS TC VII Symposium – 100 Years ISPRS, Vienna, Austria, July 5–7, 2010, IAPRS, Vol. XXXVIII, Part 7A
Contents
Author Index
Keyword Index
rg
az
az
rg
Figure 8. Part of Munich city acquired by IKONOS (upper image) and TS-X (lower image) using the accidental satellite formation.
Yellow grid lines are for better orientation between two images. Red arrows show flight (az) and look (rg) directions.
186
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertising