Generation and Accuracy Assessment of Digital

Indexed in Scopus Compendex and Geobase Elsevier, Geo-Ref Information Services-USA, List B of Scientific Journals, Poland, Directory of Research Journals www.cafetinnova.org

ISSN 0974-5904, Volume 09, No. 01

February 2016, P.P. 116-121

Generation and Accuracy Assessment of Digital Elevation Model Using Digital Photogrammetry and Differential Global Positioning System Techniques S S PANHALKAR AND AMOL P J ARAG Department of Geography, Shivaji University, Kolhapur, Maharastra, INDIA Email: [email protected] Abstract: Digital elevation model is the finest tool for visual and mathematical analysis of topography, landscapes, land forms and modeling of surface processes. Indian Cartosat-1- satellite data was designed mainly for the purpose of cartographic applications & for terrain modeling. The global availability of Cartosat-1 satellite data provides baseline information for many types of worldwide research. However, the overall generation of the DEM of this product requires additional regional as well as global studies involving ground truth control and accuracy verification methods with a higher level of precision. For the present study, part of Panchganga basin of Kolhapur district, Maharashtra has been selected. The main objective of the present study is to generate and assess the accuracy of Cartosat-1 DEM using digital Photogrammetry and DGPS techniques. Initially, block triangulation has been generated with point positioning accuracy and it is being achieved with the Rational Polynomial Coefficients (RPCs) sensor model. Thirteen GCPs were collected by using advance DGPS techniques. Interior & exterior orientations have been performed through sensor model. Finally, DEM has been generated using LPS software. RMSE & Standardized RMS statistical techniques have been applied to assess the accuracy of Cartosat-1data. Accuracy assessment reveals that Cartosat -1 satellite data is having 6.13 m and 6.23 m RMS & Standardized RMS for generated DEM. Keywords: Digital Photogrammetry, Cartosat-1, Orthorectification, DEM 1. Introduction: Digital Elevation Model plays an important role in terrain mapping. Modern Remote Sensing & Photogrammetry technology have entered in advanced digital era. The advance technology of data observing, collection, processing, storing, and production are serving for multidisciplinary studies. DEM serves as basic input for various decision making processes. Today, the relative surface products such as digital elevation model (DEM), triangular irregular network (TIN), or digital terrain model (DTM) are derived without considering ground truth dataset. These topographic data is applied to examine the nature of the terrain to aid decision making. Numerous DEM generation techniques with different accuracies for various applications have been developed. Especially with some powerful software tools, the efficient and fast DEM generation can be achieved. Thereafter, topographic details can be explored on these products. In the field of Photogrammetry, DEMs are mainly derived from image matching techniques based on the stereoscopic analysis. Today, modern photogrammetry techniques have replaced analog photographs. Digital aerial triangulation and DEM extractions are making significant development in last few years. Especially

with some powerful software tools, the efficient and fast DEM generation can be achieved. The primary objective of the present study is to generate and assess DEM of the Cartosat-1 data using a digital Photogrammetry and DGPS technique. The Cartosat-1 sensor offers high resolution in the across - track direction with 2.5 m in panchromatic mode. In addition, the optical sensor is configured with two pushbroom cameras which are mounted such that one camera is looking at +26o (band F) and the other -5o (band A) along the track. These two cameras provide stereoscopic image pairs in the same pass. Krishnasawamy et al., 2005[1] have evaluated Cartosat-1data products for terrain modeling and large-scale mapping applications. Radhika et, al., 2007 [2]; Marthaetal.,2010 [3]a;Pieczonkaetal.,2011 [4] have also used Cartosat-1 data in several other fields mainly, natural hazards assessment and estimation of hydrological parameters. Kumar et al., 2006 [5] have highlighted the processing of stereo data acquired from Cartosat-1 data to derive ortho-images and it is quite satisfactory. 1.1 Study Area: The study area (Fig. no 1) lies between 16o25' north to 16o55' north latitudes and 74o5' east to 74o30' east

#02090116 Copyright ©2016 CAFET-INNOVA TECHNICAL SOCIETY. All rights reserved.

S S P ANHALKAR AND AMOL P J ARAG

longitude. This catchment area covers part of Karveer, Hatkanangle and Shirol tahsils of Kolhapur district. The entire area of the study region is 615 sq. km. The area has diversified physiography with complex geological structure. Minimum and maximum elevation of the region is 550 m and 1020 m respectively. North western part of the catchment area is hilly with rugged topography and plain surface is towards eastern part.

117

1.4 Ground Control Point (GCPs): GCPs are points identifiable in real space. P. S. Titarov

[6] suggested that in the case of single stereo pair, four well-distributed and reliable GCPs are sufficient to achieve sub-pixel orientation accuracy. Hence, thirteen ground control points (GCPs) are collected with about 10 mm horizontal accuracy and 20 mm vertical accuracy for three stereopairs. Table 1.1: Cartosat-1 stereo coverage of the study area Sr. No.

Topo. Sheet no.

1

47 L 5

2 3

Fig 1. Location Map of Study Area 1.3 Database and Methodology: Cartosat-1 stereo images have been utilized in the present work. Three Cartosat-1 scenes were acquired from NRSC, Hyderabad of the period from 1 s t to 2 3 r d March 2007. Table 1.1 highlights the detail information of the Cartosat-1 dataset. These data sets were available in GeoTiff format with WGS84 datum reference. The methods (fig. no 2) used for DEM generation and assessment are as follows,

Extent

Path Row

160 45’ to 740 15’ to 523 318 170 00’ 740 30’ 0 0 16 30’ 74 15’ to 47 L 6 524 319 to 160 45’ 740 30’ 47 L10

160 30’ 740 30’ to 525 319 to 160 45’ 740 45’

Date of acquisition 23/3/2007 11/3/2007 01/3/2007

To collect GCPs, south DGPS instrument has been used with static surveying techniques, corrections were obtained in post processing mode. The GCPs points are acquired with UTM projection and WGS 84 datum. GCPs were placed for the roads or paths crossed walls, buildings, bridges, permanent construction etc. The GCPs were identified as such a way that they were geometrically distributed over the stereo pairs towards the four corners of the overlapped area. These collected GCPs dataset was used to refine the orientation of the images and to improve the vertical and horizontal accuracy of DEM. 1.5 New Block File Generation: The creation of DEM generation in LPS starts by creating the block file (Fig no 3) and it defines the information about the camera, block properties and proper projection. It also set frame specific information by defining the geometric models as RPC model. Cartosat-1 stereo scenes are provided with rational polynomial coefficient (RPC) within rational function (RF) sensor model. The rational polynomial coefficient (RPC) file contains the third degree polynomial coefficients that relate the image to the object space considering the imaging sensor geometry. These RPCs are sensor derived and terrain independent [7]. Rational polynomial satellite sensor models are simple empirical mathematical models and it relate image space (line and column position) to latitude, longitude, and surface elevation. The name rational polynomial derives from the fact that the model is expressed as the ratio of two cubic polynomial expressions [8]. Normalized pixel

Fig 2. Flow chart of Methodology

coordinates and normalized ground coordinates are defined by using following formula (Eq. 1 and 2).

International Journal of Earth Sciences and Engineering ISSN 0974-5904, Vol. 09, No. 01, February, 2016, pp. 116-121

Generation and Accuracy Assessment of Digital Elevation Model Using Digital Photogrammetry and Differential Global Positioning System Techniques

118 𝑙𝑁 =

𝑁𝑢𝑚 𝑙 (𝜑𝑁 ,𝜆𝑁 ,ℎ𝑁 )

𝑆𝑁 =

(1)

𝐷𝑒𝑛𝑙 (𝜑𝑁 ,𝜆𝑁 ,ℎ𝑁 ) 𝑁𝑢𝑚 𝑆 (𝜑𝑁 ,𝜆𝑁 ,ℎ𝑁 )

(2)

𝐷𝑒𝑛𝑆 (𝜑𝑁 ,𝜆𝑁 ,ℎ𝑁 )

Where, &

normalized pixel coordinates

and normalized ground coordinates h normalized height above the ellipsoid Numl, Den , Num and Den are third-order polynomials in lieu of sensor and satellite parameters. N

l

s

s

Fig 3: Block File Generation

automatically generated tie points on Cartosat images (Fig. 4). These points are lacking ground control information. By using classical point measurement tool, thirteen ground points are added to the images. The X, Y, and Z values for GCP points were provided. Triangulation process establishes relation between images, sensor model and ground points [10].

Triangulation resolves the problem of finding the position of any given point in space, if its position on two images is known [11]. This Process was completed and ground coordinates were established for all generated tie points and RMSE report was generated. However, the RMSE error was more than one pixel (1.67). Triangulation report (Fig. 5) have been checked to find out tie points which are responsible for much higher error. Those points are adjusted or removed. After editing the tie points, Triangulation was performed with 11 GCPs and 184 tie points and the RMSE (Fig no. 5) of the model was calculated which comes around 0.137 pixel. It indicates that the error in the model is within a pixel. The DEM was generated after achieving sub pixel model accuracy.

Fig 4: Automatic Tie Point Generation

1.6 Interior and Exterior Orientations: The LPS environment performs interior and exterior orientation of stereo pairs by extracting information from RPC file. Interior orientation defines the internal geometry of a sensor, as it existed at the time of image capture. Exterior orientation is the position and angular orientation of the sensor that captured the image. The Exterior orientation or rational elements describe part of the relationship between ground space coordinate system (x, y, z) and image space coordinate system (x, y, z) [9]. 1.7 Automated Tie Point Generation: LPS environment provides options for manual and automatic tie point generation. The LPS software works with image matching technique to find points in one image and identifying its conjugate point in the other stereo image. Image matching is a procedure to find correspondence between two images in the stereo model. It refers to the automatic identification and measurement of corresponding image points that are located in the overlapping areas of multiple images. As the procedure has been completed, software has

Fig 5: RMSE and Triangulation report 1.8 DEM Generation & DEM Editing: The epipolar images were generated after the process of triangulation. These images are kept as background in 3D environment for DEM editing process. Epipolar images are stereo pairs that are re-projected such that conjugate images have a common orientation and matching features between the images appear along a common horizontal axis. Stereo vision is possible with epipolar images. To check consistency of the mass points, these points are overlaid on these epipolar images. Mass points are the points which are irregularly distributed and used as the basic element to construct a Triangulated Irregular Network (TIN). Automatic



Terrain Extraction (ATE) module of LPS has been used for generating the TIN model.

119

much higher precision than the DEMs being tested. Therefore, to assess the accuracy of DEM, DGPS

generated thirteen well distributed points have been used (Table no. 2).

Fig 6. DEM extraction and editing Stereoscopic visualization is carried out to assess quality of DEM. The mass points that are misplaced are interpreted as wrong points. Each mass point plays a crucial role in defining the TIN surface. Terrain Editor tools of LPS software have been used to correct the distorted mass points. The LPS Terrain Editor facilitates verification, visualization and editing of DEM. During the DEM editing process (Fig no.6), mass points which were off the ground have been adjusted or densified by adding new mass points in the stereo environment. The location of each mass point was smartly selected to define important variations in the surface physiography. Breaklines are used to show changes in topography in terms of smoothness and continuity. Hard and soft breaklines are added to improve quality of DEM. Hard breaklines are used to demarcate ridges, streams and valleys and soft breaklines were used to mark roadway. 1.9 Ortho-Photo Generation: The next step is to create orthorectified images in LPS environment. The process of orthorectification removes terrain distortion. Orthorectified images are rectified from relief displacements and geometric errors. Hence, these images are considered as more accurate. The orthorectified images display objects in their real-world X, Y, and Z positions. Fig no. 7. shows orthorectified images of study area.

Fig 7: Orthorectified images These DGPS points are having horizontal and vertical accuracy up to 10 mm and 20 mm respectively. RMSE provides overall measure of accuracy with respect to Digital elevation model by comparing DEM values with check points of DGPS. Following formula (Eq. 3) of RMSE is used for this assessment. 1

𝑅𝑀𝑆𝐸 = √ ∑𝑛𝑖=1(𝑅𝐸𝐹𝑖 − 𝐷𝐸𝑀𝑖 )2 𝑛

(3)

Where, REFi is the reference elevation of ith location, DEMi is the elevation obtained from DEM for ith location, REF is the mean of the reference elevations of all locations n is the total number of sample locations.

2. Assessment of DEM:

A DEM is not always a true representation of the topography. It is a model of the earth’s surface and is subjected to errors like other spatial data [12]. Usually, DEM accuracy assessment is achieved by comparing the DEM values with reference values and concluded by standard deviation (SD) or RMSE [13, 14]. Reuter et al. 2009[15] suggested that the reference points should be evenly distributed across the area of interest and these check points should be measured to

Fig. 8: Scatter Plot of References Elevation vs DEM Image Observation Elevation RMS and Standard RMS result (Table no. 2) for thirteen ground control points shows that average RMS is 6.13 m and Standardized RMS is 6.23 m for generated DEM (Fig no. 9). It is also observed that there is significant


Generation and Accuracy Assessment of Digital Elevation Model Using Digital Photogrammetry and Differential Global Positioning System Techniques

120

linear relationship of DEM values with corresponding reference elevation (Fig. No. 8).

[5]

[6]

Fig.9: DEM of the study area

[7]

2.1 Conclusion: Cartosat stereo images with precise DGPS check points are used to generate DEM of Panchganga basin. Accuracy assessment reveals that Cartosat -1 satellite data is having 6.13 m and 6.23 m RMS & Standardized RMS errors for generated DEM and by using DGPS check points, accuracy of DEM can be enhanced. Horizontal accuracy is about 3.04 m and 3.16 m for latitude and about 3.01m and 3.14 m for longitude as per RMSE & SRMSE assessment. The analysis also reveals that horizontal accuracy of DEM is quite better as compare to vertical accuracy.

[8]

[9]

Acknowledgement: This paper is an outcome of research project funded by NRDMS, Department of Science and Technology, New Delhi. We express our sincere thanks to DST, New Delhi for providing financial support for the project.

[10]

References: [1] Krishnaswamy, M., Kalyanaraman, S., ( 2005) Indian Remote Sensing Satellite Cartosat1:Technical features and data products. http://www.gisdevelopment.net/technology/rs/ techrs023.htm [2] RadhikaV.N., Kartikeyan B., Gopala Krishna B., Chowdhury, S.,Srivastava P. K. (2007) Robuststereo image matching for space borne imagery. IEEE Transactionson Geosciences and Remote Sensing, 45: 29933000. doi: http:// dx.doi.org /10.1109 /TGRS.2007.898238. [3] Martha T. R., Kerle N., Van Westen, C.J., Jetten V., Vinod Kumar K. (2010) -Effect of sun elevation angle on DSMs derived fromCartosat1data.Photogrammetric Engineering and Remote Sensing, 76: 429-438. [4] Pieczonka T., Bolch T., Buchroithner M., (2011)Generation and evaluation of multi temporal digital terrain models of the Mt. Everest area from

[11]

different optical sensors. ISPRS Journal of Photogrammetry and Remote Sensing, 66(6): 927940. doi: http://dx.doi.Org/10.1016/j.isprsjprs.2011.07.003. Kumar P., Mathew J., Kumar S., Kudrat M., (2006) -Cartosat data utility for hydro power sites investigation and potential assessment .National Natural Resources Management System (NNRMS), Bulletin,31:61-67. P.S. Titarov., (2008) -Evaluation of cartosat 1 geometric potential. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B1. Beijing 2008 PP. 841-846. Rao, C. V., Sathyanarayana, P., Jain, D. S., Manjunath, A. S., (2007) -Topographic map updation using Cartosat-1 data. In Proceedings of the 2007 Annual Conference of the Remote Sensing & Photogrammetry Society (RSPSoc2007)), Newcastle upon Tyne.National Remote Sensing Agency. Gopala Krishna , B., Amitabh, Srinivasan, T. .P, Srivastava, P. K.., (2008) -DEM generation from high resolution multi-view data product. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences Vol. XXXVII. Part B1. pp 1099-1102. Grodecki J., Dial G., (2003) -Block adjustment of high resolution satellite images described by rational polynomials. Photogrammetric Engineering& Remote Sensing, 69 (1): 59-68. Krishna Murthy et,al., (2008) -Analysis of DEM generated using Cartosat-1 stereo data over Mausanne Les Alpiles – Cartosat scientific appraisal programme (CSAP TS – 5). The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B1. Beijing 2008 PP. 13431348. Hartley, R and Sturm, P., (1997) –Traingulation, computer vision and image understanding, 68(2): pp14-157.

[12] Wechsler S., (2006) -Uncertainties associated

with digital elevation models for hydrologic applications: a review. Hydrology and Earth System Sciences, 3: 2343-2384. doi: http://dx.doi.org/10.5194/hessd-3-2343-2006. [13] Nikolakopoulos K.G., Kamaratakis E.K., Chrysoulakis N., (2006) -SRTM vs. ASTER elevation products comparison for two regions in Crete, Greece. International Journal of Remote Sening, 27: 4819-4838. doi: http://dx.doi.org/10.1080/01431160600835853 . [14] Tsutsui K., Rokugawa S., Nakagawa H., Miyazaki S., Cheng C.T., Shiraishi T., Yang



S.D., (2007) -Detection and volume estimation of large-scale landslides based on elevationchange analysis using DEMs extracted from high-resolution satellite stereo imagery. IEEE Transactions on Geoscience and Remote

Sensing,

121

45

(6):

1681-1696.

doi:

http://dx.doi.org/10.1109/TGRS.2007.895209 . [15] Reuter, H.I., Hengl, T., Gessler, P., Soille, P., (2009) -Preparation of DEMs for geomorphometric analysis. In: Hengl, T., Reuter, H.I. (Eds.), Geomorphometry: Concepts, Software, Applications. Elsevier, Oxford, pp. 87e120.

Table no: 2 DGPS and DEM image evaluation using GCPs Sr.No Latitude 1 2 3 4 5 6 7 8 9 10 11 12 13

1857197.45 1842603.65 1831664.17 1852938.75 1830092.75 1842869.13 1844882.75 1850489.97 1853785.86 1831721.88 1831147.04 1829223.73 1828229.62 Sr.No 1 2 3 4 5 6 7 8 9 10 11 12 13

DGPS Observation Longitude

Height

Latitude

401676.63 411390.27 397368.39 423236.34 419670.43 430767.40 450132.32 458405.07 442262.06 427086.47 440684.94 441093.22 457903.71

502.88 501.00 500.65 496.08 472.06 562.23 577.10 570.12 647.89 559.73 548.44 561.58 573.50

1857198.65 1842606.35 1831660.42 1852940.66 1830095.8 1842871.23 1844879.95 1850486.87 1853790.67 1831724.58 1831143.4 1829219.44 1828228.72

Latitude Errors (m) Difference Square -1.2 1.44 -2.7 7.29 3.75 14.06 -1.91 3.64 -3.05 9.3 -2.1 4.41 2.8 7.84 3.1 9.61 -4.81 23.13 -2.7 7.29 3.64 13.24 4.29 18.4 0.9 0.81 DX RMSE (m) 3.04403578 SRMSE (m) 3.168765

Longitude Errors (m) Difference Square -0.7 0.49 -3.28 10.75 -2.03 4.12 -3.88 15.05 3.93 15.44 -3.48 12.11 2.46 6.05 3.98 15.84 2.64 6.96 3.6 12.96 2.3 5.29 -2.33 5.42 -2.8 7.84 DY 3.01687561 3.140416

DEM image Observation Longitude 401677.33 411393.55 397370.42 423240.22 419666.5 430770.88 450129.86 458401.09 442259.42 427082.87 440682.64 441095.55 457906.51

Height 509.43 494.44 503.67 500.35 480.56 555.44 580.66 579.22 644.59 564.83 540.84 556.18 579.6

Height (m) Difference Square -6.55 42.9 6.56 43.03 -3.02 9.12 -4.27 18.23 -8.5 72.75 6.79 46.1 -3.56 12.67 -9.1 82.81 3.3 10.89 -5.1 26.01 7.6 57.76 5.4 29.16 -6.1 37.21 DZ 6.1308802 6.238908
