Institute of Photogrammetry and GeoInformation Research Current projects
Geometric & Semantic Analysis of Space Imagery for Topographic Mapping & Database Construction of Selected Experimental Test Area in Turkey (2005)

Geometric & Semantic Analysis of Space Imagery for Topographic Mapping & Database Construction of Selected Experimental Test Area in Turkey (2005)

Team:  Karsten Jacobsen
Year:  2005
Funding:  Scientific and Technical Research Council of Turkey (TÜBİTAK), International Bureau Jülich Research Center Germany
Duration:  2002-2005
Is Finished:  yes

Research Group: Geometric aspects of sensors and images

Contact Person: Karsten Jacobsen

sponsered by:

Scientific and Technical Research Council of Turkey (TÜBİTAK)  

International Bureau  Jülich Research Center Germany

Topic:
Optimizing update and creation of small and medium scale maps including digital elevation models and creation of ortho-images especially based on space images. 

Background:
The geometric model of most of the imaging systems to be used, has been evaluated. Special attention was taken to the not published mathematical model of IKONOS images. The information contents of most of the space images to be used has been investigated, but not in the same area, what is required for a sufficient comparison and not under the special conditions of Turkey.  This has to respect also the usual information contents of the Turkish topographic maps. The map contents is different in the different countries.

Specific Objectives of the Project:
Special attention has to be taken into the optimized use of the different types of space images for mapping. An economic solution has to be found respecting all steps of procession including image costs, required control information, DEM generation, creation of ortho-images and mapping including the identification of the required objects. Not in any case the highest resolution space images, which may be expensive and limited in the covered area, have to be used.

Methodology to be followed:
The geometric relation between image and object space has to be determined by bundle orientation. For the line scanner images the Hannover University program BLASPO and for the perspective image the program system BLUH can be used. This includes also the determination of systematic image errors by self-calibration with additional parameters. For the IKONOS-Geo-images and level 1B-data, the program CORIKON was developed, which is independent upon rational functions. Based on the image orientation, digital elevation models shall be and partially have been determined by automatic image matching, followed by a qualified filtering of the digital surface to a digital elevation model, for this the programs DPCOR and RASCOR can be used. The extraction of map information can be based on a stereo evaluation or a registration based on ortho-images by mono-plotting. Both methods shall be investigated.
The geometric accuracy of the derived maps will be compared with existing maps, but also some independent check points. The semantic information will be compared with existing maps, differences have to be analyzed for the reason, which may be caused by a change of the object, which has to be used as map update, errors of existing maps, but also misidentifications and not recognizing of objects.

International Cooperation:
The work will be done in a cooperation between the Zonguldak Karaelmas University and University of Hannover. The University of Hannover will especially take care about the geometric relation and will support this also by required extension of the used software. The Zonguldak Karaelmas University will take and took care about the required control points, the reference maps and the extraction of map data

Expected Outputs of the Project:
The project will give an overview about the potential of the available high resolution space images usable for mapping and map update. This will include also estimations of the required time for data acquisition and the economic aspects. Existing gaps in the software chain will be closed. The knowledge about mapping based on space images will be established, this is also an important preparation for the use of the images of the planned Turkish microsatellite BiltenSAT.

Expected Benefits for both Countries:
Experiences with optimal conditions for mapping will be achieved. This improves the possibility for an actual map contents in Turkey which is a fundamental condition for the development of the country. For the German counterpart also the knowledge and software chain will be improved. Also in Germany the results of these investigations may be used to optimize the map update of small scale maps. The survey administrations have already had the first view to this topic. The results of the project will be published.

TURKISH PROJECT TEAM  

Y.Doç.Dr. Gürcan Büyüksalih ( project coordinator)
Y.Doç.Dr. Hakan Akçın
Arş. Gör. Güven Koçak
Uzman Murat Oruç
Uzman M.Aycan Marangoz
GERMAN PROJECT TEAM

Dr.-Ing. Karsten Jacobsen (project coordinator)
Prof.Dr.-Ing. Christian Heipke
Dipl.-Ing. Helge Wegmann
Dipl.-Ing. Ursula Wissmann
Dipl.-Ing. Adelheid Elmhorst
Zonguldak Karaelmas University
Dept. of Geodesy & Photogrammetry
67100 Zonguldak / Turkey
Institute for Photogrammetry and GeoInformation (IPI)
University of Hannover
Nienburger Str.1
D-30167 Hannover

First results:
Generation of digital elevation model with ASTER-images
A stereo model of ASTER-images (pixel size 15m) has been used for the generation of a digital elevation model (DEM). The striping of the original images (level 1A) could be removed with the specially developed mini-program RADCOR. The image orientation was possible with the Hannover program BLASPO with root mean square errors at 18 control points with: RMSX=+/- 11.4m, RMSY=+/- 10.2m and RMSZ=+/- 14.6m. The control points have been digitized from topographic maps 1 : 50 000, by this reason no better accuracy can be expected – the discrepancies are dominated by the map quality.

  orientation of   ASTER-model:

  distribution of control points

-  vertical discrepancies shown as red vectors

-  horizontal discrepancies shown in black

  The upper left part is covered by the Black Sea

 

 

 

By automatic image matching with program DPCOR corresponding image points have been generated by least squares matching. The high quality of the near infrared ASTER channels allowed a matching of 84% of the points with a correlation coefficient exceeding 0.80

gray value coded DEM determined with ASTER images by DPCOR and BLASPO gray value image generated with LISA BASIC

  3D-view to the DEM based on ASTER-images

A test of the DEM against independent check points, collected from maps 1 : 25 000, showed mean square differences in the range of SZ=+/-12m, which is a realistic result for the pixel size of 15m and a height to base relation of 1.7. Against points measured by GPS on roads, mean square differences in the height of approximately 16m have been shown, but these values are influenced by steep slopes just beside the roads in the mountainous area.

Orientation of IKONOS-image

Two IKONOS Geo-images are available for the same area of Zonguldak taken in September and November 2002. The Geo-image is a simple rectification to a horizontal plane chosen by SpaceImaging. In the mountainous area of Zonguldak a rectification of the image with a nadir angle of 17.6° is leading to mean square discrepancies at the 39 control points determined by GPS of +/- 22m and for a nadir angle of 30° to +/-42m – a not sufficient result for the ground pixel size of 1m. With the Hannover program CORICON the Geo-images can be improved based on three-dimensional control points without use of the rational functions from Space Imageing, just based on the nominal collection elevation and azimuth as approximate values. So a totally sufficient accuracy for the first scene of RMSEX=+/- 1.15m and RMSEY=+/- 1.17m and for the second scene of RMSEX=+/- 1.18m and RMSEY=+/- 1.05m could be reached, that means just a little more than 1 pixel.

A comparison of the discrepancies at the control points of both scenes shows a strong correlation of larger vectors. Because of the different view directions, this only can be explained by discrepancies of the control point coordinates itself and / or the point identification in the different images.

 
discrepancies at control points in transformed IKONOS-Geo-image –
 2 different scenes with different view directions, but same control points

 

8.0In the test area a KVR-1000 orthoimage, created by Sojuzkarta with a pixel size of 1.56m, is also available. The comparison with check points in the same coordinate system showed not acceptable shifts of 108m in X and 89m in Y. After shift correction the mean square discrepancies are in the range of 15m showing a rotation of the orthoimage. A similarity transformation reduced the root mean square errors to +/-9m with correlations of neighbored check points. Based on a least squares interpolation (prediction), the discrepancies could be reduced to a local fit of +/-3.4m.

First Tests of the TK-350-image geometry
The Russian space photograph TK-350 has an image size of approximately 300mm x 450mm. An image deformation may happen, by this reason the reseau crosses of the TK-350 with a spacing of 10mm have been measured at the analytical plotter Zeiss Planicomp P1 and in the digital image with the Hannover program DPCOR. The images have been scanned with a cartographic scanner from Purup because the photogrammetric scanner do not allow the scan of such an image size. An analysis of the reseau crosses has been made at first with the P1-measurements against the nominal grid coordinates and with the digital measurements against the P1-measurements to separate the different error sourc

RMS differences

RMSx RMSy'

RMS systematic
components
RMS random
component
P1 against nominal value
Image 324
8.0 8.3 5.2 5.0 6.1 6.6
Digital data against P1
Image 324
10.7 9.7 7.3 6.4 7.8 7.2
Digital data against nominal values Image 326 7.5 7.8 3.7 4.6 6.5 6.3
Digital data against P1
Image 326
6.2 7.7 2.4 4.1 5.7 6.5

Root mean square differences TK-350 reseau crosses [µm]

The systematic discrepancies between the Planicomp measurements and the nominal reseau positions do show the image deformation, which is only slightly larger like in aerial images. The random part shows local image deformations, which are also limited. The comparison of the reseau positions in the digital image, scanned with a pixel size of 17 µm, against the location in the photo, measured at the Planicomp shows the accuracy of the pointing in the digital image and the accuracy of the used image scanner. The systematic differences of 7.3µm, 6.4µm, 2.4µm and 4.1µm are confirming the astonishing accuracy of the cartographic scanner. The random errors are in the range of 0.3 up to 0.4 pixels. So no general geometric problems are existing.

Conditions for Mapping
For the planemetric representation a stereoscopic capability is not required if the height information is available. An orthoimage can be used for mono plotting, but nevertheless a stereoscopic view is supporting the object identification.
As mentioned, as rule of thumb, the pixel size on the ground shall not exceed 0.05 up to 0.1mm in the map scale, corresponding to 1.25 up to 2.5m for mapping in the scale 1 : 25 000. If this condition cannot be matched, not the whole required map contents can be extracted and a more intensive field check is required. The range between the lower and the upper value of this rule is depending upon the structure of the area, the national map standards and the imaging conditions. Not in any case, the information contents is directly depending upon the pixel size – colour information improves the object identification and sometimes the image quality is affected by atmospheric conditions.
The reason for the dependency of the mapping scale upon the pixel size is based on the fact, that smaller map scales do not include so many details. For example in a map scale 1 : 5000 all building corners are shown, while in a map scale 1: 100 000 the city centre is only showing building blocks.

 
Figure 1: information contents of different map scale
left: original information          right: reduced, comprehended and simplified contents for smaller map scale


Figure 2: comparison of information contents 
 left: JERS  L-Band                                                   right: Landsat 7 band 
  18m pixel size                                                         30m pixel size 

Also the type of image is important for the information contents like obviously in figure 2. The identification of objects in a synthetic aperture radar (SAR) image like JERS cannot be compared with the information contents of an optical image. In addition to the general difficulty of object identification is a SAR-image quite depending upon the view direction. In addition we do have the geometric problems of foreshortening, layover and shadows. By these reasons SAR-images are only used for mapping in areas with more or less permanent cloud coverage and for special topics like determination of flooded areas. This situation may change with the very high resolution SAR-images like from TerraSAR-X and SAR-X Cosmo Skymed which shall be launched in 2005 and shall have a pixel size of 1m. Nevertheless also this pixel size should not be compared with a 1m-pixel-size from an optical sensor. The main advantage of SAR is more located in the generation of digitalelevation models (DEM) by interferometric SAR (InSAR).

  
 Figure 3:  
    Landsat 7 panchromatic,                                            Terra ASTER,
    15m pixel size                                                                15m pixel size

Landsat 7 includes also a panchromatic band with 15m pixel size. This pixel size is not far away from the useful size of 10m, but the quality of the panchromatic Landsat images is still limited. The contrast is not optimal and so not many more details can be identified like in a colour composite of different Landsat bands. The Terra ASTER image shown in figure 2 on right hand side allows the identification of quite more objects. Of course the colour is supporting this, but also the original image quality is still better. But corresponding to the mentioned relation between pixel size and possible map scale, these images do not allow the generation of maps 1 : 100 000.


Figure 4:
    TK 350, 10m pixel size                                         IRS-1C, 5.8m pixel size

With a higher resolution more details are visible in the images. IRS-1C always shows nearly the complete road structure, building blocks and large individual buildings like required for the map scale 1 : 50 000


Figure 5:
    IKONOS multispectral                                                 KVR-1000   photo,
    4m pixel size                                                              1.6 m pixel size


Figure 6:
    IKONOS panchromatic,
    1m pixel size

The advantage of the higher resolution becomes clear with the figures 4 and 5. The panchromatic IKONOS image shows every building and even cars can be seen. Also the KVR-1000 photo shows a sufficient information contents for maps 1 : 25 000. The grain of the photo is still visible, but it does not disturb so much. The colour of the multispectral IKONOS-images makes the classification of objects more simple, but with 4m pixel size not the same details can be seen like with 1m pixel size. The advantage of the colour can be combined with the high resolution of the panchromatic image by a combination to a so called pan-sharpening, which is usually based on an IHS-transformation, followed by a geometric fit of the lower resolution multispectral image to the panchromatic image, an exchange of the intensity channel of the IHS-image by the panchromatic image and a back transformation from IHS to RGB.
Not in any case the image quality is the same, the atmospheric condition may have a negative influence and the sun-elevation is also important. By this reason in some cases the results of mapping will not be the same like under optimal conditions.
The preceding shown space images of the area around Zonguldak do show very well the dependency of the information contents about the pixel size, but also the type of imaging.


relation - possible map scale and pixel size of used image