Advanced
Generation of Simulated Geospatial Images from Global Elevation Model and SPOT Ortho-Image
Generation of Simulated Geospatial Images from Global Elevation Model and SPOT Ortho-Image
Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography. 2014. May, 32(3): 217-223
Copyright © 2014, Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http:// creativecommons.org/licenses/by-nc/3.0) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
  • Received : April 04, 2014
  • Accepted : June 06, 2014
  • Published : May 27, 2014
Download
PDF
e-PUB
PubReader
PPT
Export by style
Share
Article
Author
Metrics
Cited by
TagCloud
About the Authors
Wan Yong Park
Member, Agency for Defense Development (E-mail:wypark@add.re.kr)
Yang Dam Eo
Corresponding Author, Member, Division of Interdisciplinary Studies, Dept. of Advanced Technology Fusion, Konkuk University (E-mail:eoandrew@konkuk.ac.kr)
Abstract
With precise sensor position, attitude element, and imaging resolution, a simulated geospatial image can be generated. In this study, a satellite image is simulated using SPOT ortho-image and global elevation data, and the geometric similarity between original and simulated images is analyzed. Using a SPOT panchromatic image and high-density elevation data from a 1/5K digital topographic map data an ortho-image with 10-meter resolution was produced. The simulated image was then generated by exterior orientation parameters and global elevation data (SRTM1, GDEM2). Experimental results showed that (1) the agreement of the image simulation between pixel location from the SRTM1/GDEM2 and high-resolution elevation data is above 99% within one pixel; (2) SRTM1 is closer than GDEM2 to high-resolution elevation data; (3) the location of error occurrence is caused by the elevation difference of topographical objects between high-density elevation data generated from the Digital Terrain Model (DTM) and Digital Surface Model (DSM)-based global elevation data. Error occurrences were typically found at river boundaries, in urban areas, and in forests. In conclusion, this study showed that global elevation data are of practical use in generating simulated images with 10-meter resolution.
Keywords
1. Introduction
The acquisition of optical images has become more difficult due to the small number of observation days that meet photography conditions and an increase in airborne debris and smog that are caused by the expansion of urbanization and industrialization. It has been difficult to generate seamless image maps under the same weather and climate conditions because of the increase in the length of time needed to acquire images that compose the complete maps (Park, 2000).
Satellite or airborne cameras acquire geospatial images. A geospatial image includes the terrain and artificial features on the terrain. Simulated geospatial images refer to images that would be acquired by an observation sensor that stores random sensor positions and attitude elements. These images are simulated, using conventional aerial/ satellite images, DSM data, and acquisition conditions (Yun ., 2002). These simulated images are used to restore images that were blocked by clouds or other obstacles, by integrating them with the actual images. They are also useful for simulating images that are visualized under particular seasonal or meteorological environments in a flight simulator (Seidel and Datcu, 1993). In addition, while monitoring a wide region, the geometrical and environmental conditions of cameras are simulated to observe particular sitjavascript:;uations more closely, and the results are used in determining optimal flight tracks (Stark, 1993; Kim et al ., 2013).
Studies have been made on the generation of simulated low-resolution environmental satellite and SAR images. Visible and near-infrared images of small areas have been simulated for military use. Simulated optical images are generated based on a Digital Elevation Model (DEM) and geospatial images, using ray-tracing technique. DEM accuracy is then assessed by calculating self-consistency between the original DEM and the image that is generated from the stereo-simulated images. During the process, a quantitative assessment was made on whether the simulated images were regenerated in the form of the original image, and the effectiveness of the simulation was evaluated (Woo, 2013; Schott et al ., 1995).
This study attempted to generate geometrical simulated images that are the basis for the generation of simulated geospatial images. For the pixel values of the generated images, new values are allocated, considering various characteristics of reflectance radiance and atmospheric correction that are dealt with in general simulated images. In this study, using SPOT-3 panchromatic images, ground control points, and elevation data at 5-meter grid intervals, ortho-images were generated through sensor modeling under collinearity conditions (Stark, 1993). For the simulated images of 10-meter spatial resolution, exterior orientation parameterbased ephemeris data were used. Ortho-images, global elevation data at 30-meter grid intervals GDEM2, and SRTM1 DEMs were adopted.
This process was used to verify accuracy in the generation of the images that were simulated by the difference of spatial resolution between elevation data used at generation of ortho-images and data used at generation of simulated images. Lastly, to analyze the difference in position accuracy (by geometric pixel) between original images and simulated images obtained using global elevation data, a similarity examination was conducted by comparing the image coordinates, which were obtained through the collinearity condition equation by using elevation data.
2. Generation of Simulated Optical Images Under Collinearity Conditions
Simulated images are generated by using exterior orientation parameters, ortho-images and elevation data, based on user-desired spatial resolution, with the assumption that images are obtained by viewing the ground at the position and in the posture designated by the user. When original images (prior to geometric correction) are used in the generation of simulated images, ground control points should be applied to the images to establish a sensor model. On the other hand, in ortho-images, because ground coordinates are already known, it is unnecessary to include the error intervention factors for users' coordinate computation. Simulated images can be generated in a costefficient manner by using conventional geospatial images.
Ortho-images and elevation data have an effect on the quality of simulated images. With respect to coregistration between elevation data and ortho-images, and from a theoretical standpoint, the use of elevation data in generating original ortho-images is best for the generation and co-generation of simulated images. If the ephemeris data of the original images used for the generation of orthoimages are unknown, the ortho-images can be assumed to be close to true images. Better results are expected as the elevation data become more precise.
In general, to generate various types of simulated images, it is necessary to acquire DEM and ortho-images that have higher resolution than the images to be generated. In reality, this is not easy. In this study, a simulated imagegeneration test was performed with global elevation data and easy-to-access ortho-satellite images. The geometrical reproductivity of the simulated images was then assessed.
Fig. 1 depicts the concept of the generation of simulated images under collinearity conditions.
  • ① Identify image coordinateslandcon the area of the images to be generated (AOI).
  • ② Estimate the 2D ground coordinatesXandYon the identified image coordinateslandc, using orientation parameters and back projection under collinearity conditions. The elevation value (ℎ) is estimated by applying the ray-tracing technique to the point where LOS and DEM are crossed.
  • ③ Identify image coordinates at the same spot from the already estimated ground coordinatesXandYin the ortho-images.
  • ④ Extract the pixel values on the coordinates (landc) of the simulated images from the coordinates of the identified ortho-images and allocate them.
PPT Slide
Lager Image
Concept of generation of simulated images under collinearity conditions
Repeat steps ① through ④ for the entire AOI of the simulated images.
If collinearity conditions are used, it is necessary to follow a relatively complicated process, which includes observing ground control points and calculating exterior orientation parameters. In addition, some process should be handled by experts. This study focused on analyzing the geometrical reproductivity of simulated images. The order of image generation was original image, ortho-image, and simulated image. Finally, the original images were examined to determine whether they match the simulated images.
3. Test Materials and Generation of Simulated Images
- 3.1 Test area and materials
The selected test areas shown in Fig. 2 were in Iksan, Jeollabuk-do. To test the geometrical reproductivity of simulated images, the land-cover type in the test areas had to be equal to the diverse elevation distribution, assuring that the results could be generalized. In the test areas, urban land, crop land, undeveloped land, water, and forest were properly distributed with an average elevation of 48 meters (ranging from 0 to 501 meters).
PPT Slide
Lager Image
Test area and SPOT image
The SPOT-3 satellite images used for the test were 10-meter resolution panchromatic images that were obtained on April 5, 1995, at the viewing angle of 8.6°. The coordinates at the ground control point were obtained manually, using an analytical stereoplotter. The highdensity elevation data at 5-meter grid intervals were produced based on the digital topographic map at a scale of 1:5,000, to generate ortho-images from the original images. To generate simulated images, global elevation data at 30-meter grid intervals, GDEM2 and SRTM1 were used. These data are shown in Fig. 3 . For the datum and coordinate systems of the elevation data, the WGS84 and UTM systems, respectively, were adopted.
PPT Slide
Lager Image
Types of elevation data for the experiment
- 3.2 Generation of simulated images
- 3.2.1 Generation of ortho-images
The exterior orientation parameters to be used for image simulation are determined based on SPOT images and ground control points. A bundle adjustment under collinearity conditions was used for the modeling method. In addition, ERDAS LPS (Leica Photogrammetry Suit) 2010 was used to determine exterior orientation parameters and to generate ortho-images. A total of 12 ground control points were used for the bundle adjustment. The modeling accuracy standards are 'Less than 1 pixel.'
To generate simulated images based on the 10-meter resolution SPOT images, the high-density elevation data at 5-meter grid intervals were produced, based on the digital topographic map at 1:5,000 scale, and bilinear interpolation was adopted. WGS84 and UTM were selected for the datum and coordinate systems, respectively. In general, orthoimages and elevation data that will be used for the future generation of simulated images were used. Therefore, the use of general commercial tools for the generation of orthoimages meets the objectives of the test.
- 3.2.2 Generation of simulated images
Simulated images with 10-meter spatial resolution were generated by using the ortho-images and global elevation data at 30-meter grid intervals GDEM2 and SRTM1 DEMs, respectively, based on the ephemeris data from exterior orientation parameters. This process was targeted to check the possibility of generating simulated images according to differences in the spatial resolution of elevation data and to examine their accuracy.
To generate simulated images and assess their geometrical accuracy, a commercial tool (Mathworks' MATLAB-based program) was used. Simulated images are generated by using exterior orientation parameters, ortho-images, and elevation data, to reproduce original images with ephemeris data under collinearity conditions. Therefore, the geometrical quality of the simulated images is determined by the accuracy and precision of raw data. Eqs. (1) and (2) present the back projection of SPOT images under collinearity conditions, which were used in determining the pixel values of the simulated images:
PPT Slide
Lager Image
PPT Slide
Lager Image
where, ( X0,Y0 ) : the location of the perspective center, h : the height of the ground (Xp,Yp) , xp and yp : image coordinates, x0 and y0 : principal point transition f : focal length, Xp, Yp, Zp : ground coordinates, m : rotation matrix. To the pixel values of the coordinates ( xp, yp ) of the sumulated image in other words, the pixel values of the ground coordinates ( Xp, Yp ) of the ortho-image are allocated. The pixel values of all image coordinates in the area where simulated images are generated are estimated in the same manner.
- 3.2.3 Analysis of simulated images
To analyze the pixel location accuracy of original images and the simulated images that are generated using global elevation data, the variances of simulated images and image coordinates calculated under collinearity conditions were estimated. Geometrical similarity was then examined. Since image coordinates and ground coordinates are calculated in a real number during the generation of original images, ortho-images, and simulated images, the pixel values of original images are altered by interpolation. Therefore, index-based approaches with conventional image similarity have limitations as a similarity index for this study (Ji and Gallo, 2006). Therefore, to quantificationally calculate accuracy according to geometric reproduction between original and simulated images, this study compared the pixel coordinates that are calculated by using SRTM1 and GDEM2 elevation values under collinearity conditions with values estimated by using high-density elevation data. Eqs. (3) and (4) are collinearity-condition equations for calculating image coordinates:
PPT Slide
Lager Image
PPT Slide
Lager Image
According to Eqs. (3) and (4), the elevation values ( Zp ) applied to the calculation of the coordinates of original images were applied to SRTM1 and GDEM2 elevation values. Position changes were then examined by the difference between the coordinates of original images that were calculated with high-density elevation data, to determine if they matched the original images. It appears that a large gap between the two images occurs because of differences between high-density elevation data used for the generation of ortho-images from original images, and SRTM1 and GDEM2 elevation data that are used in simulating original images from ortho-images.
4. Result Analysis
Ortho-images with 10-meter resolution were generated by using exterior orientation parameters and high-density elevation data from the SPOT original images. Fig. 4 shows the boundary of Iksan City, a target area for the generation of ortho-images and simulated images.
PPT Slide
Lager Image
Iksan city boundary (in red) on the SPOT orthoimage
The simulated images generated from the ortho-images using the global elevation data, SRTM1 and GDEM2 DEMs were reproduced at the same pixel intervals with SPOT original images.
Figs. 5 and 6 show differences between high-density elevation data used in generating ortho-images and the SRTM1 and GDEM2 elevation data that were used for the generation of simulated images. The algorithm in this paper presents the back projection of SPOT images under collinearity conditions, which were used in determining the pixel values of the simulated images. The height ( h ), which applied to algorithms, should be determined at the point where LOS meets the surface as shown in Fig. 7. Therefore, the geometrical displacement has resulted from the DEM difference. When compared (by the naked eye) to the highdensity elevation data, the difference was larger in the GDEM2 elevation data than in the SRTM1 elevation data.
PPT Slide
Lager Image
Distribution of height difference between highdensity DEM and SRTM1
PPT Slide
Lager Image
Distribution of height difference between highdensity DEM and GDEM-2
PPT Slide
Lager Image
Geometrical displacement by DEM difference
The results are quantitatively compared in Table 1 , below.
Height difference between high density DEM and global elevation model (unit: m)
PPT Slide
Lager Image
Height difference between high density DEM and global elevation model (unit: m)
Table 1 shows a difference between elevation data used for the generation of ortho-images and elevation data used in generating simulated images. Therefore, a collinearity condition equation was applied to quantitatively examine the geometrical displacement between original and simulated images. Table 2 shows the results.
Geometrical displacement of simulated image by height difference (unit: pixel)
PPT Slide
Lager Image
Geometrical displacement of simulated image by height difference (unit: pixel)
The elevation difference shown in Table 2 takes place when calculating the pixel difference with 10-meter resolution, based on the viewing angle (8.6°) of the SPOT images. The SPOT image used for the experiment was acquired under the viewing angle of 8.6°, and 10-meter ground-sample distance (GSD). Therefore, the horizontal error of 1 pixel size (10m) can be computed to a vertical error, as follows:
PPT Slide
Lager Image
According to the test, the displacement of the simulated images generated, based on SRTM1 and GDEM2,was 'less than 1 pixel' by 99.99%. Figs. 8 and 9 identify errorreported areas by displaying the distribution of unmatched pixels (0.2 pixel or higher for geometrical displacement) for a visual inspection.
PPT Slide
Lager Image
Distribution of displacement (> 0.2 pixel) in simulated images (SRTM1)
PPT Slide
Lager Image
Distribution of displacement (> 0.2 pixel) in simulated images (GDEM2)
Figs. 8 and 9 show that the location of error occurrences in simulated images was fairly similar in both cases, which means the SRTM1 and GDEM2 elevation data has a similar accuracy. However, the locations of error occurrences result from the elevation difference of topographical objects and high-density elevation data generated from the DTM and DSM-based global elevation data. This was often found at river boundaries, in urban areas, and in forests.
Table 3 shows that the pixel displacement (at precision levels) of SRTM1 simulated images are lower than GDEM- 2 simulated images, which means SRTM-1 is superior in elevation similarity. At precision levels 'less than 1 pixel', pixels were included by 99.99% regardless of the type of base data. When compared to the test results of NGA in the U.S., which comparatively analyzed elevation using SRTM and GDEM elevation data, the results of this study seem feasible (Tachikawa et al ., 2011). Experiment results show that global elevation data are practical in generating 10-meter simulated images.
Geometrical displacement and percentage of simulated image pixels by height difference
PPT Slide
Lager Image
Geometrical displacement and percentage of simulated image pixels by height difference
5. Conclusion
In this study, the similarity of open ortho-images and simulated images (which were generated using elevation data) to original images were assessed. When global elevation data related to the Korean Peninsula (in which inaccessible areas were included) were used, the error consistency of SRTM1 in the generation of simulated images was superior. In the generation of simulated images, in particular, position changes within 1 pixel were over 99%, which confirms practicality. Test results showed that, when compared to high-density elevation data, global elevation data SRTM1 and GDEM2 seemed sufficiently accurate. When 10-meter resolution and maximum observation angle (±27°) of SPOT satellite were considered, a difference of about 20meters in elevation is included in a 1-pixel displacement. This confirmed that it was sufficient to generate simulated images, along with the already developed ortho-images. In this study, spectral radiance values were generated and used for the pixels of geometrical simulated images. Therefore, we expect that more practical simulated images can be generated. Because of easy access, this study applied the generation of simulated images under collinearity conditions to the test. To improve user convenience, an orbital model-based program of simulated image generation is planned for future study.
Acknowledgements
This research was supported by the Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education (NRF-2013R1A1A2007582).
References
Ji L. , Gallo K. 2006 An agreement coefficient for image comparison Photogrammetric Engineering & Remote Sensing 72 (7) 823 - 833
Kim S. , Eo Y. , Lee B. , Jang I. , Han S 2013 Remote sensed image simulation methodology Journal of Next Generation Information Technology(JNIT) 4 (8) 111 - 117
Park G. 2000 Analysis of the Sensor Characteristics and Imaging Time for Mapping, TEDC-409-000570 Agency for Defense Development Daejeon, Korea
Schott J. , Salvaggio C. , Brown S. , Rose R. 1995 Incorporation of texture in multispectral synthetic image generation tools SPIE 2469 189 - 196
Seidel K. , Datcu M. 1993 Fusion of real and synthetic images for remote sensing scene understanding FRACTAL ‘93 Kingston, UK 1 - 12
Stark R. 1993 Synthetic Image Generator Model:Application of View Angle Dependent Reflectivity Components and Performance Evaluation in the Visible Region, Master's thesis Science in the Center for Imaging Science, Colleage of Imaging Arts and Sciences of the Rochester Institute of Technology New York, USA 241 -
Tachikawa T. , Kaku M. , Iwasaki A. , Gesch D. , Omioen M. , Zhang Z. , Danielson J. , Kreiger T. , Curtis B. , Haase J. , Abrams M. , Crippen R. , Carabaijal C. 2011 ASTER Global Digital Elevation Model Version 2- Summary of Validation Results ASTER GDEM Validation Team Japan and USA 1 - 27
Woo Dong-Min 2013 Quantitative assessment of 3D reconstruction procedure using stereo matching Journal of IKEEE 17 (1) 001 - 009
Yun Y. , Cho W. , Park J. , Lee J. 2002 A Study on the generation of simulated high-resolution satellite images Korean Journal of Remote Sensing 18 (6) 327 - 336