1. 34 American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) ISSN (Print) 2313-4410, ISSN (Online) 2313-4402 Β© Global Society of Scientific Research and Researchers http://asrjetsjournal.org/ Evaluation of Mapping Accuracy of High-Resolution Stereoscopic Satellite Images Hossam H. El-Semary* Faculty of Engineering-Shobra , Benha University-Cairo-Egypt Email: hossam.elsemary@feng.bu.edu.eg Abstract High resolution satellite images is still used in large scale mapping due to the need to produce fast products. High resolution stereoscopic satellite images present good enough 3d products that include the benefits of large-scale coverage and low-cost products. A stereopair of IKONOS satellite is used in this research that covers a part of North Sudan country. The study handles the 3d mapping accuracy of using stereoscopic satellite images. The study gives a spotlight on the accuracy in X, Y, Z and the space vector R. Another view of this study the N, E and elevation is indicated. The research environment is mainly ENVI software due to its capabilities of topographic processing module. Some distributed set of ground points (control and tie) was determined on the images and then observed using GPS surveying. Several experiments have been performed to evaluate the resulted mapping product. Keywords: Stereoscopic; High Resolution; Satellite Images; Spatial Accuracy. 1. Introduction Many satellites provide imagery with high impacts in various geomatics fields. Using stereoscopic satellite images with a good base over height ratio can produce a good Digital Elevation Model (DEM). On can mention IKONOS that has been successfully used for 3d mapping and objects extraction. If orthoimages are also generated or available, textured 3D models can be created and used for photo-realistic visualization [1]. Photogrammetric technique is an appropriate solution for obtaining the DEM of large areas. DEM data can easily be obtained using stereo images through photogrammetric methods [2]. The main objectives of this study are: To evaluate the mapping accuracy using high resolution stereoscopic satellite image. In addition; to generate the final DEM resulting from this system. ------------------------------------------------------------------------ * Corresponding author. http://asrjetsjournal.org/ American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2019) Volume 58, No 1, pp 34-43 35 2. Research Data Stereoscopic products consist of two overlapped satellite images of a location on Earth, taken from two different perspectives position on the same track. IKONOS stereo pair contains an image collected at elevation angle above 60 degrees as well as an image collected at a higher elevation angle above 72 degrees with 30Β°-45Β° convergence with 0.54 to 0.83 base-to-height ratio. Figure 1 shows IKONOS image data collecting system with in track sampling. Figure 1: Stereo IKONOS Satellite Image data collection. Reference Stereo products have a horizontal accuracy of 25 meters CE90 (Circular Error 90%) and a vertical accuracy of 22 meters LE90 (Linear Error of 90%) without any Ground Control Points (GCPs). When reliable GPS derived GCP is available the horizontal and vertical Geospatial accuracies increase to <2.5m horizontal and <1.5m vertical [3]. The test area (Wadi Seidna) is located at the north of Khartoum in Sudan bout 16km north of Omdurman city and geographically, it lies between upper left corner coordinates of (444535.977, 1747921.762) and lower right corner coordinates of (449800.083, 1742045.500). The elevation range of this area is about 60 m. The study area is about 31 Square Kilometers. Figures (2) and (3) show the general location of the study area. Table 1 indicates the relative polynomial coefficients for the stereo pair that can be used for image rectification model. http://www.satimagingcorp.com/satellite-sensors/ikonos.html American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2019) Volume 58, No 1, pp 34-43 36 Figure 2: General location of the study area. (a) Left Image (b) Right Image Figure 3: IKONOS HRS 1M stereo-pair of the study area. American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2019) Volume 58, No 1, pp 34-43 37 Table 1: Main parameters of the available IKONOS HRS 1M stereo-pair. Acquisition Date/Time 08:41 GMT 27-12-2003 and 08:42 GMT 27-12-2003 Sun Angle Azimuth 156.1241 and156.3809 degrees Sun Angle Elevation 63.50707 and 47.50945 degrees Overlap 99% Rows 5893 and 6004 pixels Columns 5351 and 5357 pixels Pixel Size X 1.000 meters Pixel Size Y 1.000 meters Percent Component Cloud Cover 0 Size (km) 5.808 km * 5.344 km 3. Methodology The developed system depends on input, output, processing and decision blocks and can be represented by flowcharting [4]. Figure 4 indicates the block diagram of the developed system where it is starting by input both left and right images to Envi (as a processing software). The GCPs must entered to carry out the rectification model and space resection by determining the stereopair parameters required for space intersection after. Then the space coordinates of the check (tie) points is calculated so accuracy assessment can be carried out by comparing its measured GPS coordinates with the calculated values. If the quality of the results is quite good the process can be stopped or a change in GCPs and tie points must be carried out. Figure 4: Block diagram of the research work. American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2019) Volume 58, No 1, pp 34-43 38 The change may include elimination one or more GCPs according to how much it is far from the mathematical model (absolute error). If the result is accepted, then DEM of the study area can be generated using the same software of Envi. A lot of trials were made to reach the optimum case. Generation of DEM is a minor objective, so no accuracy assessment is established. Since the IKONOS satellite imagery vendor, Space Imaging Company, has not released the satellite ephemeris data, no physical mathematical model can be established. So, the most common applied mathematical model is the rational function model (RFM). The Rational Function Model (RFM) is a generalized model that is widely used. RFM has come into widespread use within the mapping community. The related research showed that both RFM and the polynomial model can reach reasonably satisfying accuracy. The RFM expresses each of the x and y image coordinates as a ratio of two polynomial functions. It can be represented as follows: π‘₯ = 𝑃1(𝑋,π‘Œ,𝑍) 𝑃2(𝑋,π‘Œ,𝑍) = βˆ‘ βˆ‘ βˆ‘ π‘Žπ‘–π‘—π‘˜π‘‹π‘–π‘Œπ‘–π‘π‘–π‘š3 π‘˜=0 π‘š2 𝑗=0 π‘š1 𝑖=0 βˆ‘ βˆ‘ βˆ‘ π‘π‘–π‘—π‘˜ π‘‹π‘–π‘Œπ‘–π‘π‘–π‘›3 π‘˜=0 𝑛2 𝑗=0 𝑛1 𝑖=0 (1) 𝑦 = 𝑃1(𝑋,π‘Œ,𝑍) 𝑃2(𝑋,π‘Œ,𝑍) = βˆ‘ βˆ‘ βˆ‘ π‘π‘–π‘—π‘˜π‘‹π‘–π‘Œπ‘–π‘π‘–π‘š3 π‘˜=0 π‘š2 𝑗=0 π‘š1 𝑖=0 βˆ‘ βˆ‘ βˆ‘ π‘‘π‘–π‘—π‘˜π‘‹π‘–π‘Œπ‘–π‘π‘–π‘›3 π‘˜=0 𝑛2 𝑗=0 𝑛1 𝑖=0 (2) Where, x,y Image coordinates X,Y,Z Ground coordinates ai jk, bi jk,cijk,di jk Polynomial coefficients (total 80) m1,m2,m3,n1,n2,n3 0-3 , wherei+j+k ≀ 3 The polynomial coefficients are called rational function coefficients (RFCs). The RPCs are calculated by Space Imaging from the satellite ephemeris and attitude data instead of releasing the ephemeris data themselves. Therefore, the RFM model is implemented by most of the commercial software packages to use the supplied RPCs. This software modules deal with IKONOS satellite images by reading the RPC files and applying the RFM model to orient the IKONOS imageries. Static relative GPS positioning technique was used for measuring the ground control points. Two well identified points of the High Accuracy Reference Network provided by the Sudanese Survey Authority, near the study area, were chosen as reference stations. Ten out of twenty-two points were selected to be ground control points and the other twelve points were used as check points. Figure 4 shows the whole set of the measured ground points while Figure 5 shows the distribution of control and check points. Geo-referencing and stereo-model setup of the images was established using different numbers and different distribution of GCPs until get the best accuracy of 10 ground control points. The 10 pyramid points represent the Ground Control Points, while the rest of the points (rounded points are used as check points). Point 1 lies on the border of the stereopair so it is marked with red pyramid in figure 4. American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2019) Volume 58, No 1, pp 34-43 39 Figure 5: Ground control points and check points. The final GPS coordinates of ground control and check points are listed in table 2 in the format of easting, northing and height (E, N, H). The red color is used for control and green color is used for check points. Figure 6: Selected Ground control points & check points. American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2019) Volume 58, No 1, pp 34-43 40 Table 2: Final GPS coordinates for both GCPs and check points. NO ID Type Easting Northing Height 1 GCP28 Control 449548.020 1747432.638 381.723 2 GCP29 Control 444593.142 1747670.157 404.44 3 GCP27 Control 449694.505 1742120.183 384.672 4 GCP43 Control 447409.125 1742295.875 388.773 5 GCP44 Control 446865.860 1745528.241 402.499 6 GCP45 Control 449610.442 1744352.352 386.826 7 GCP46 Control 446593.192 1745203.946 405.533 8 GCP48 Control 448312.423 1742333.837 386.439 9 GCP49 Control 447552.933 1747529.038 391.878 10 GCP50 Control 449268.7255 1742087.394 385.938 11 CP30 Check 447336.227 1747394.633 393.181 12 CP33 Check 446924.385 1747321.87 394.619 13 CP34 Check 447381.562 1744428.317 436.551 14 CP36 Check 448561.519 1746118.83 390.771 15 CP35 Check 447557.070 1746175.505 395.371 16 CP37 Check 449603.553 1742438.988 383.786 17 CP38 Check 449472.709 1743051.137 383.910 18 CP39 Check 449685.294 1742977.897 382.444 19 CP41 Check 449088.085 1742137.513 385.269 20 CP47 Check 447846.782 1742200.025 390.105 21 CP32 Check 446886.200 1747091.386 395.575 22 CP42 Check 449171.625 1743275.145 385.820 To apply the mapping accuracy assessment; the following equations is used by computing the RMS error in easting, northing directions and height respectively. 𝑅𝑀𝑆𝑒𝐸 = √ 1 𝑛 βˆ‘ βˆ†πΈπ‘–2𝑛 𝑖=1 (3) 𝑅𝑀𝑆𝑒𝑁 = √ 1 𝑛 βˆ‘ βˆ†π‘π‘–2𝑛 𝑖=1 (4) 𝑅𝑀𝑆𝑒𝑍 = √ 1 𝑛 βˆ‘ βˆ†π‘π‘–2𝑛 𝑖=1 (5) RMSeE = Root Mean Square error in E direction RMSeN = Root Mean Square error in N direction American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2019) Volume 58, No 1, pp 34-43 41 RMSeZ = Root Mean Square error in Z ΔΕі= the X Residual for GCP Ρ– ΔΝі= the Y Residual for GCP Ρ– Ξ”ZΡ–= the Z Residual for GCP Ρ– Ρ– = GCP Number n = the number of GCP 4. Results and Analysis Figure (5) shows the final rectified layered map produced using manual selection to guarantee the high performance. Ten points were chosen to testify the performance of the mapping accuracy. Table (3) indicates the results of the selected testing ground points. The average absolute error is computed and found equal to 0.5814 m in E while it was 0.5369 m in N and equals 0.8372m in total planimetric. This result is near 1.5 ground pixel size of the satellite image. Another way to indicate the results is the standard deviation which was 0.333287624 m in E while it was 0.226860508 m in N and 0.294663469 m in total planimetric. This result is near 0.5 ground pixel size of the satellite image. Table 3: results of the selected ground points. NO Name E map(m) N map(m) E GPS(m) N GPS(m) RE(m) RN(m) TOTAL(m) 1 GCP 11 449684.000 1742978.340 449685.295 1742977.897 -1.295 0.443 1.368 2 GCP 12 447553.505 1747528.516 447552.933 1747529.038 0.572 -0.522 0.774 3 GCP 13 449695.200 1742119.531 449694.505 1742120.183 0.695 -0.652 0.952 4 GCP 14 446593.784 1745204.606 446593.192 1745203.946 0.592 0.660 0.886 5 GCP 15 449548.480 1747432.320 449548.020 1747432.638 0.460 -0.318 0.559 6 GCP 16 444592.960 1747669.730 444593.142 1747670.157 -0.182 -0.427 0.464 7 GCP 17 448312.860 1742333.630 448312.423 1742333.837 0.437 -0.207 0.483 8 GCP 18 446865.990 1745527.170 446865.86 1745528.241 0.129 -1.071 1.078 9 GCP 19 447847.237 1742199.598 447846.78 1742200.025 0.455 -0.427 0.623 10 GCP 20 449611.439 1744351.71 449610.44 1744352.352 0.997 -0.642 1.185 American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2019) Volume 58, No 1, pp 34-43 42 Figure 7: Final Rectified layered map for the study area. 5. Conclusion One can conclude that the accuracy of the mapping product using stereo satellite images can reach to 1.5 of ground pixel size so the user of this type of products can take this information in his consideration. Using the standard deviation in this study gives another way to understand the accuracy limits of mapping process from stereo satellite images. One can conclude that 0.5 standard deviation of the ground pixel size is the accuracy limits for this type of products. In general one can conclude that it is good to make mapping using stereoscopic satellite images in geodetic and topographic mapping products but it is not convenient to use it in cadastral mapping products 6. Recommendations It’s necessary to recommend using stereoscopic satellite images in the production of small scaled mapping products (topographic or geodetic maps). It is not recommended to use stereoscopic satellite images in cadastral mapping due to the lack of accuracy comparing with unmanned aerial vehicles (UAVs). One American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2019) Volume 58, No 1, pp 34-43 43 must recommend the comparison between stereoscopic satellite imagery and RADAR imagery. References [1]. A. Serwa, Automatic Extraction of Topographic Features from Digital Images. PhD thesis, Cairo: Faculty of Engineering in Cairo- Azhar University, 2009. [2]. Tridipa Biswas, Kamal Pandey, Annadurai R., "Satellite Photogrammetry based DEM Generation using Satellite Stereo Pair Images and Terrain Parameters Extraction," Int. Journal of Advances in Remote Sensing and GIS,, vol. 4, no. 2, pp. 64-76, 2016. [3]. S. I. Corporation, "IKONOS Stereo Satellite Imagery," 19 July 2012. [Online]. Available: www.satimagingcorp.com.svc/ikonos-stereo-satellite-image.html. [4]. A. Serwa and H. Semarry, "Integration of Soft Computational Simulator and Strapdown Inertial Navigation System for Aerial Surveying Project Planning," Spatial Information Research (SPIR), vol. 24, no. 3, p. 279–290, June 2016. [5]. A. Serwa, "Development of Soft Computational Simulator for Aerial Imagery Project Planning," Surveying and Land Information Science (SaLIS), vol. 75, no. 2, 2016. [6]. K. Murthy, M. Shearn, B. D. Smiley, A. H. Chau, J. Levine and D. and Robinson, "Skysat-1: very high- resolution imagery from a small satellite.," International Society for Optics and Photonics, 2014. [7]. Michael Xie, Neal Jean, Marshall Burke, David Lobell, and Stefano Ermon, "Transfer Learning from Deep Features for Remote Sensing and Poverty Mapping," in AAAI Conference on Artificial Intelligence (AAAI-16), 2016. [8]. M. Jordan, "Why the logistic function? A tutorial discussion on probabilities and neural networks," Massachusetts Institute of Technology, Massachusetts, 1995. [9]. B. DasGupta and G. Schnitger, the Power of Approximating: A Comparison of Activation, Advances in Neura Information Processing Systems ed., San Mateo, CA: Morgan Kaufmann Publishers, 1993. [10]. C. Γ–zkan and F. S. Erbek, "The Comparison of Activation Functions for," Photogrammetric Engineering & Remote Sensing, vol. 69, no. 11, p. 1225–1234, 2003. [11]. Wang, Qian; Wang, Yi; Niu, Ruiqing; Peng, Ling, "Integration of Information Theory, K- Means Cluster Analysis and the Logistic Regression Model for Landslide Susceptibility Mapping in the Three Gorges Area, China," Remote Sensing, vol. 9, no. 938, p. 9, 2017. [12]. Charlotte Pelletier, Silvia Valero, Jordi Inglada, Nicolas Champion, Claire Marais Sicre and GΓ©rard Dedieu, "Effect of Training Class Label Noise on Classification Performances for Land Cover Mapping with Satellite Image Time Series," Remote Sensing, vol. 9, no. 2, p. 2, 2017. [13]. Ratna Sari Dewi, Wietske Bijker, Alfred Stein and Muh Aris Marfai, "Transferability and Upscaling of Fuzzy Classification for Shoreline Change over 30 Years," Remote Sensing, vol. 10, no. 9, p. 1377, 2018.