1031-1040_GS-003.qxd 8/12/10 2:39 PM Page 1031

New Era of Cartosat Satellites for Large Scale Mapping

P.V. Radhadevi, S.S.Solanki, V.Nagasubramanian, Archana Mahapatra, D. Sudheer Reddy, M.V. Jyothi, Krishna Sumanth, J. Saibaba, and Geeta Varadan

Abstract (Baltsavias et al., 2001; Jacobsen, 2003; Grodecki and Dial, Important considerations for large scale mapping from 2003; Fraser et al., 2002; Radhadevi et al., 2008), automatic satellite images are information content and geometric DTM/DSM generation (Jacobsen, 2004; Toutin, 2004; Poli et al., fidelity. Cartosat series of satellites with stereo mapping 2004). Comparison of information contents of high-resolution capability have become the mainstay towards large scale space images for the purpose of mapping is explained by mapping for urban and rural applications. Algorithms for Topan et al. (2004). The Indian space program witnessed processing of high-resolution Indian remote sensing satellite several major accomplishments and scaled newer heights in data has been developed at ADRIN and is used for opera- mastering space technology during the last few years. The tional generation of data products. Variations in the sensor Cartosat series of satellites with stereo mapping capabilities model with respect to the viewing geometries of Cartosat-1 have become technically suitable for large scale mapping for and Cartosat-2 are explained in the paper. Finally, an urban and rural applications. A method for strip processing assessment of the mapping potential of the satellites is the Cartosat-1 data is explained in Srivastava et al. (2008). discussed. The geometric accuracy achieved from Cartosat-1 In our work, the geometric model is based on the and Cartosat-2 images over the same checkpoints are com- viewing geometry of the satellite, combining the principles pared. DEM, geometric accuracy, and capability for topo- of photogrammetric collinearity equations, originally devel- graphic feature capture are good enough for making 1:10 000 oped for SPOT-1 (Radhadevi et al., 1994), and further adapted and 1:7 000 scale maps from Cartosat-1 and Cartosat-2, and tested for different sensor geometries from IRS-1C/1D to respectively. Based on the error estimation and analysis, it is Cartosat-2. The sensor position, velocity, and attitude is concluded that if the strict photogrammetric processing model derived from the given supplementary data (ADIF: Ancillary and ground control points are employed, high-resolution Data Information File format), and its variations are modeled satellite imagery can be used for the generation and update of using a simple polynomial model. The aim of this paper is topographic maps of scale 1:10 000 and larger. to bring out the correction methodologies for processing a stable and an agile satellite along with a comparative study on the mapping potential of the Cartosat series of satellites. Introduction Technical details for Bundle Block Adjustment (BBA) of The launch of high-resolution satellites, such as Cartosat-1 Multi-view images of Cartosat-2, DEM and Ortho-image and Cartosat-2 are revolutionizing the field of digital map- Generation and Matching of Ortho-images are explained. ping in India, particularly in urban areas where satellite imagery has rarely been used in the past. With the increased awareness of applications of high-resolution satellite imagery Features of Cartosat-1 and Cartosat-2 to solve local land administrative problems, more and more Cartosat-1 (IRS-P5) is the first satellite of ISRO designed to state governments are preferring high-resolution satellite provide high-resolution, along-track stereo imagery for mapping applications. The platform contains two panchro- imagery over existing methods. For the full exploitation of ϩ Ϫ the potential of this data, the “classical” satellite image matic camera payloads with 26° and 5° tilted with processing methods must be extended in order to describe respect to nadir. The base to height ratio is about 0.62. the imaging geometry. In general, the processing of these Stereo acquisition geometry of Cartosat-1 is shown in kinds of images provides a challenge for algorithmic Figure 1. Data is quantized with 10 bits and integration redesign, and this opens the possibility to reconsider and time is 0.336 ms with nominal GSD of 2.5 m. Each CCD has improve many photogrammetric processing components, 12,000 pixels, separated in to 6,000 each of odd and even such as image enhancement, image orientation (georeferenc- pixels. These odd and even pixel rows are separated by 35 ␮m (equal to five pixels in image plane). The staggered ing), ortho-rectification, DTM/DSM generation, and object extraction. Many different geometric models of varying array configuration of the cameras is shown in Figure 2. complexity, rigor and accuracy have been developed for Large swath, high-resolution and stable imaging configura- rectification of satellite images. In recent years, a large tions are the key features of Cartosat-1, which make it amount of research has been devoted globally to efficiently suitable for large scale mapping. utilize these high spatial resolution imagery data. Examples can be found in sensor modeling and image orientation Photogrammetric Engineering & Remote Sensing Vol. 76, No. 9, September 2010, pp. 1031–1040. Advanced data Processing Research Institute, Department 0099-1112/10/7609–1031/$3.00/0 of Space, Manovikasnagar P.O., Secunderabad -500 009, © 2010 American Society for Photogrammetry India ([email protected]). and Remote Sensing

PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING September 2010 1031 1031-1040_GS-003.qxd 8/12/10 2:39 PM Page 1032

Figure 3. Paint-brush mode of acquisition of Cartosat-2.

Figure 1. Stereo acquisition for Cartosat- 1 (source: ISRO).

images of Cartosat-2, high-resolution DEMs can be produced. There are two CCD arrays in the payload: main array and redundant array, both having a 12,000 pixels array, similar to Cartosat-1. Several new technology elements like a high performance star sensor, improved Inertial Reference Unit, dual-gimbaled antenna, high-bit telemetry and data handling systems, light weight, and compact spacecraft structure have been introduced into the design of Cartosat-2. The precise ephemeris and attitude data allows for reducing the number of ground control points. Furthermore, this information enables direct georeferencing of the imagery without geomet- ric reconstruction of the imaging process and ground control. The data from the satellite is used for detailed mapping and other cartographic applications at the cadastral level, urban and rural infrastructure development and management, as well as applications in Land Information System and Geo- graphic Information Systems. More details about Cartosat-1 and Cartosat-2 payload systems are available in Krishnaswami (2002). Figure 2. Staggered array configuration. Preprocessing As illustrated in Figure 2, the optical detectors on the odd and even arrays are not arranged in a simple matrix of rows and columns on ground. The payload design of Cartosat-1 India’s highest resolution imaging satellite, Cartosat-2, and Cartosat-2 is such that the odd and even detectors are was launched in 2007. The advancement in GSD for Cartosat-2 staggered by five scan lines (35 ␮m) in the focal plane. is achieved with “Step and Stare” (SNS) mode of acquisition. However, this separation in the focal plane does not trans- Cartosat-2 is an advanced remote sensing satellite capable of late into a constant stagger in the image data. The impact of providing user requested scene-specific targeted imagery. line separation in the focal plane during imaging with With the panchromatic camera (PAN) on board, the satellite different viewing configurations is analyzed. Stagger varies can provide imagery with a spatial resolution of better than along the track with different scan angles which results in a one meter and a swath of 9.6 km on each overpass. In variable sampling pattern on ground. Therefore, video “SPOT” mode of imaging, the length of the strip can be as alignment by sliding the images from the two arrays (by five long as 290 km. If a programmed area of interest is wider lines) is not feasible. The stagger parameters are computed than 9.6 km, more than one swath is required to capture the by the reconstruction of the viewing geometry with a imagery. However, as the Cartosat-2 sensor can swing in any calibrated camera geometry model. Reconstruction of the direction, it can capture up to four swaths (covering an area viewing geometry includes the exterior and interior orienta- of up to 38 km wide) in one overpass or on one day. This is tions of the sensor. The image data of both staggered lines how the paint-brush mode of acquisition of Cartosat-2 is can be combined using a one-dimensional re-sampling exercised to increase swath in single overpass. The paint- which accounts for all these effects. The model is designed brush acquisition mode of Cartosat-2 is shown in Figure 3. to provide an accurate method of transforming points from The satellite can be steered up to 45° along as well as across image space to object space and vice versa. As Cartosat-2 is track, which make it a highly maneuverable satellite mission. more agile and its scan angle is continuously changing Because of the agility of the satellite, multi-view imagery also during acquisition, stagger correction and video alignment is can be acquired over the same area for production of digital a major preprocessing consideration for Cartosat-2 compared elevation models (DEMs). Three views; nadir, fore, and aft are to Cartosat-1. Figure 4 shows stagger value computed over a captured to form stereo triplets. By processing multi-view strip of Cartosat-2.

1032 September 2010 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING 1031-1040_GS-003.qxd 8/12/10 2:39 PM Page 1033

Ϫ where ( f, xs, ys) are the image coordinates, (X, Y, Z) are the coordinates of the object point, (Xp, Yp, Zp) is the perspective center d is the scale factor, and M is the rotation matrix, i.e., Ϫ Ϫ f m00 m01 m02 X Xp ϭ Ϫ xs d. m10 m11 m12 . Y Yp . (2) Ϫ ys m20 m21 m22 Z Zp P Q P Q P Q After rearrangement, Equation 2 can be written as two equations: ϭ ϩ B Ϫ B ϭ f1 (fm10 xsm00)(X Xp) 0 (3)

ϭ ϩ B Ϫ B ϭ f2 (fm20 ysm00)(X Xp) 0 (4) ϭ ϭ where m00 (m00, m01, m02), m10 (m10, m11, m12), Figure 4. Along-track differences between odd ϭ m20 (m20, m21, m22), etc. and even arrays computed over a Cartosat-2 ϭ strip of 290 km. Here, rotation matrix M QGI * QIO * QOB * QBP * QPC where, QPC is the CCD to Payload transformation matrix, QBP is the payload to body transformation matrix, QOB is the Body to Orbit matrix, QIO is the Orbit to Inertial matrix, and QGI is the Inertial to ECEF matrix. Ground coordinates are computed for a particular pixel Combined Camera Adjustment for Cartosat-1 in an odd array using the initial fitted trajectory. Height The challenge of developing a combined camera model for at this particular point is interpolated from a public Cartosat-1 is to find common parameters for both images or domain DEM like SRTM. Re-projection of this ground point to establish their relative orientation. The benefit is that the is done to the even array. The difference between the number of unknown parameters is reduced which in turn image co-ordinates computed in the even array and odd will also gain a reduction of the correlation between the array gives the effective stagger parameter. unknown parameters. The developed model is very flexible. Rigorous Sensor Model (RSM) In the sensor model, two along track images can be used and treated as a whole. The unknown parameters of both A sensor model reconstructs the imaging geometry. Recon- images are computed together. The most important require- struction of the viewing geometry includes the exterior and ment is to understand the satellite motion, to represent this interior orientations of the sensor. The model relates 3D motion using a set of equations, and derive from these object point position to their corresponding 2D image equations, the state vectors and attitude parameters at a positions by describing the geometric relationship between specific point of the trajectory. In a tied camera approach, the image space and object space. The rigorous models in-flight calibration of the cameras is a very important attempt to describe the physical properties of the sensors prerequisite. The objective of in-flight calibration is to acquisition and are based on collinearity equations, which compute the individual and relative alignment offsets of the are extended in order to describe the specific geometry of cameras and include them in the sensor model (in the push-broom sensors. The algorithm for orbit attitude model Payload to Body transformation matrix). Remaining errors at presented in this paper, which combines the principles of the system level are due to small uncertainties in the body viewing geometry of the satellite with photogrammetric attitude, which are shared by both the cameras. This can be collinearity equations, was originally developed for SPOT-1 corrected with few conjugate points. Each conjugate point and IRS-1C/D. This model has been adapted to suit the serves as two GCPs separated in the time line by 52 seconds camera characteristics of Cartosat-1 and Cartosat-2. The which can be used for correction of a long orbital arc. information about of the satellite ephemeris, attitude, Indirectly, a conjugate point used for correction serves as a sidereal angle, and time is extracted from the ADIF. The orbit tie-point between Aft and Fore images, and the adjustment parameters in the collinearity equations are position (X , Y , p p becomes a stereo strip triangulation. A generic polynomial Z ), velocity (V , V , V ) and attitude parameters are roll (␻), p x y z model is developed so that by selecting the order of polyno- pitch (␾), and yaw (␬). Satellite position and orientation are mial, it can be adapted for different types of sensors. This given at every 125 msec. Attitude data in terms of quater- option allows the modeling of the sensor position and nion are converted into Euler angles. For predicting position attitude with 3rd, 2nd or 1st order polynomials. Corrections and attitude parameters at desired times from the given up to constant and 1st order coefficients are done for telemetry data, a polynomial curve fit is made. The initial Cartosat-1. Over a long pass, the variation in the attitude values of all the parameters are derived by least squares angles will not be a bias, and therefore, time-dependent adjustment to the ephemeris data using a generalized coefficients are also updated with GCPs. polynomial model. Collinearity equations express the fundamental relation- ship that the perspective center, image point and object Bundle Block Adjustment of Multi-view Images of Cartosat-2 point lies on the straight line, i.e., Cartosat-2 is an agile satellite. Figure 5 shows the attitude Ϫ Ϫ variations over a single long strip (SPOT mode of imaging). f X Xp For multi-view images, the strip length will be compara- ϭ Ϫ xs d.M. Y Yp (1) tively less with respect to SPOT mode of imaging. The Ϫ ys Z Zp bundle adjustment approach for photogrammetric point determination with three images acquired from Cartosat-2 is P Q P Q

PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING September 2010 1033 1031-1040_GS-003.qxd 8/12/10 2:39 PM Page 1034

for each of them are very different. Then, the matched points are transformed to the raw space. A subset of the total number of points is selected as tie-points. Corrections to the imaging event parameters and the tie-point ground coordinates are computed after the bundle block adjustment.

DEM and Ortho-image Generation DEM and ortho-image generation is done using an Iterative Ortho-image Refinement (IOR) method. In IOR, we start from a coarse DEM and generate ortho-image and refined DEM in an iterative way using a stereo pair. The quality and availability of SRTM data worldwide has benefited the DEM user community at large. However, the digital eleva- tion model for non-European and non-US countries is available only at an interval of 3 arc seconds. This posting is not suitable for many applications. The method of IOR utilizes the capability of Cartosat images and availability of SRTM data to generate and simultaneously edit good quality DEM. The main advantage of this process is to utilize the inherent geometric relationship between the digital elevation model and ortho-image to automatically Figure 5. Attitude variations over the strip of Cartosat-2. edit the generated DEM. Since the matching is performed between the ortho-images, the success rate and reliability is much better. For a given DEM (SRTM) and exact exterior orientation parameters, the ortho-images generated from left and right images are supposed to be identical. Since our existing DEM a least-squares adjustment based on the collinearity may not be so accurate, they may not co-register. The equations (Equation1) modified for a block. mismatches are attributed to inaccurate-height in DEM. The Equations 3 and 4 after modifying can be written as: error in height is estimated using the mismatch and the updated DEM. The corrected DEM is used to generate another ϭ s # i ϩ # i # q Ϫ q i pair of ortho-images and again the mismatch is computed. f1 xij M0 f M1 (Xj Xp) (5) a b The major steps are regular DEM generation, establishing a relation between object and image space, generation of f ϭ ys # M0i ϩ f # M2i # (Xq Ϫ Xq i ) (6) ortho-images, matching of ortho-images, and computation of 2 a ij b j p height error. This process is repeated until the mismatch is reduced to a defined threshold. Output of IOR is a B i ϭ i i i refined/improved DEM and ortho-images. In the refinement where, Xp (Xp, Xp,Xp) represents the coordinates of B th ϭ process, one of the important assumptions is to have the satellite position vector for i imaging event, Xj (Xj,Xj,Xj) knowledge of exact exterior orientation parameters. Since a s s represents the ground coordinates of point j, and (xi j,yi j) is combined camera adjustment model is used for Cartosat-1 the photographic coordinates of the image point; jth point in and Bundle Block Adjustment for Cartosat-2 multi-view the ith image: images, the uncertainty in the knowledge of exterior orienta- tion parameters is a minimum between the cameras/images. M0i ϭ mi , mi , mi , M1i ϭ mi , mi , mi , Thus, the IOR method for DEM and ortho-image generation c 00 01 02 d c 10 11 12 d becomes more meaningful.

M2i ϭ mi , mi , mi etc. Matching of Ortho-images c 20 21 22 d The ortho-images are matched using the area-based correla- tion technique in hierarchical mode. Ortho-images are All tie-point measurements as well as all available generated first at certain pixel size. For example, if the control point coordinates are handled simultaneously in one required DEM posting is 100 m and selected ortho-resolution single adjustment process which gives the guarantee of high is 20 m, every fifth pixel of the ortho-images will be precision results. Even based on the same image coordinates, matched. Image pyramids are generated, and points are an independent model adjustment cannot reach the same matched at lower levels. The lower level matched points are quality; this is due to the data reduction by relative orienta- transferred to the next level. These points are used to tion, the comparatively inexact handling of systematic image predict the approximate conjugate position and cross- errors, and the usual separate computation of the horizontal correlation is used to find the correct position. To have and the vertical unknowns. greater confidence, matching is done left to right and In order to process the block of three images (multi- followed by right to left. Through intersection at the view), it is necessary to identify the tie-points between matched points, we can compute the position of a particular them. Tie-point matching process in all the three images is point. Height error is computed by the equation: done simultaneously because all the three strips overlap. Feature-based matching is done between the images after ¢h ϭ ¢r * (B/H) (7) generating the system level geocoded products from all the three. Using a Forstner operator, candidate tie-points are where ⌬h is the height error to be added with existing identified first. Matching between the raw images is not height, and ⌬r is the mismatch difference in terms of ground recommended because the orientation and attitude angles co-ordinates.

1034 September 2010 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING 1031-1040_GS-003.qxd 8/12/10 2:39 PM Page 1035

Comparison of Mapping Aspects of Cartosat-1 and Cartosat-2 Data Important criteria or considerations for large scale map- ping from satellite images are (a) information content, (b) geometric fidelity, and (c) radiometric quality. Informa- tion content or the feature delineation will be increased as and when the spatial and spectral resolution of the satellite improves. But achievable geometric accuracy depends upon many factors including the stability of the satellite, method of imaging, control points used, and the model used to rectify the imagery. A pilot study was conducted to evaluate the mapping potential of Cartosat-1 and Cartosat-2 satellites. Satellite orientation parameters are updated using the correction methodologies explained in the previous sections. GCP requirement is minimized by using the ephemeris and attitude data with appropriate weights for adjustment and precise payload geometry in to the adjustment model. Figure 6. Accuracies from Cartosat-1. Dataset Details To check the achievable accuracies from Cartosat-1, we have used the base maps as reference data which satisfy 70 cm accuracy. Point-based accuracy evaluation is done over six datasets. Area covered in each data is 30 km ϫ 30 km. Dataset 1 is over the Delhi area; height range over this area is approximately 200 to 300 m. Dataset 2 covers Cochin city which is a coastal area where the height range is from 0 to 50 m. Dataset 3 is over Hyderabad area which is a mixed terrain with height range of 500 to 700 m. Data set 5 and 6 are over the Alwar region where the height variation is from 200 to 800 m. Fifty distributed check- points are identified in the Fore and Aft images for each dataset. Apart from these datasets, DEM accuracy was evaluated over a dataset on Himalayan region for Cartosat- 1 where the height range is 4,000 to 6,000 m. A 1:50 000 scale map generated by Survey of India is used in this region as the source for GCP. Cartosat-2 multi-view images are available only over the first three areas (over Delhi, Cochin, and Hyderabad). Each dataset for Cartosat-2 contains three views, Fore, Nadir, and Aft. Automatic tie- Figure 7. Accuracies from Cartosat-2. points are identified between them. Fifty distributed check points are identified manually in the reference base maps and the corresponding image locations in the Cartosat-2 images for accuracy evaluation. Only a single GCP is used for orientation of the sensor in all the cases for Catosat-1 and two GCPs for Cartosat-2.

Accuracy Evaluation Errors obtained for Cartosat-1 and Cartosat-2 are given in Figures 6 and 7. It is clear from Figure 6 and Figure 7 that the achiev- able accuracies from Cartosat-1 data are of the order of 3 to 4 m with just one GCP, and corresponding accuracies for Cartosat-2 are of the order 2.0 to 2.2 m with just two GCPs. Evaluation is done at independent checkpoints using the sensor model, and also on a generated raster DEM with reference data. DEM and ortho-images are generated from the data sets and the accuracy is evalu- ated. Having the mass points of correlation from the matching process as well as the exterior and interior orientation of the camera system, the surface heights can be calculated using forward intersection. This is done by least squares adjustment for the intersection of the image rays. Figure 8 shows the contours generated from the DEM over an area where height range is about 4,000 to 6,000 m. Figure 8. Contours shown on the Cartosat-1 ortho-image The “kinks” in the contours are exactly falling over the draped over DEM. ridges of actual terrain. The quality of DEM is also evident

PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING September 2010 1035 1031-1040_GS-003.qxd 8/12/10 2:39 PM Page 1036

Figure 9. Vector layer from reference data superimposed onto the spatially accurate ortho-images over the same area: (a) Cartosat-2 image, (b) Cartosat-1 image.

from the contours which are generated without any towards mapping. Figure 12 shows a comparison of geomet- filtering. ric accuracy achieved from Cartosat-1 and Cartosat-2 images Ortho-images are generated from Cartosat-1 and Car- over the same set of checkpoints. It is obvious from this tosat-2 covering the Cochin area and the vector layers from figure that achievable accuracy will not linearly improve in the reference data are superimposed over the products in the same proportion as resolution increases. Both satellites, original resolution as shown in Figure 9. All road networks Cartosat-1 and Cartosat-2, show almost same range of and building corners extracted from the reference image are accuracies with same set of checkpoints. RMS errors in along exactly co-registered with the image at the corresponding track direction for Cartosat-1 and Cartosat-2 are 1.7 m and locations. DEM and ortho images are generated from Cartosat- 1.3 m, respectively, as shown in Figure 12a, and RMS errors 1 and Cartosat-2 datasets mentioned, Figure 10 displays the in the across track direction are 2.4 m and 1.7 m, respec- results with Dataset 5 (Alwar) of Cartosat-1. tively, as shown in Figure 12b. DEM is generated from Dataset 3 (Hyderabad) is shown in Figure 11. The radiometric resolution of both Cartosat-1 Capability of Topographic Data Capture and Cartosat-2 is 10-bits. The agility of Cartosat-2 requires a The ortho-rectified imagery was analyzed with the capture proper function for point spread correction and image of spatial information. Figure 13 the portion of ortho- restoration; this is an important preprocessing requirement. images from Cartosat-1 and Cartosat-2 at their original Another concern of the high view angles of Cartosat-2 is the resolutions. effect of shadows over built-up areas which degrades the From the ortho-images generated over the test areas, mapping capability. The DEM and Ortho-image generated sub-regions are analyzed to ensure the capability of topo- from Cartosat-2 are shown in Figure 11. graphic data capture, that is, urban, semi-urban, and rural. In order to make a comparison, the same datasets were In each of these areas we attempted to capture all the processed for Cartosat-1 and Cartosat-2. Keeping the source features present according to the specifications at various of GCP same, both were compared for error analysis for same mapping scales. The features collected in this study set of checkpoints and with same set of GCP’s (used for included roads, railway tracks, paths, buildings, vegetation correction). Thirty-three points were manually identified in limits, water features, and field boundaries. Many of the the common regions over Delhi acquired from Cartosat-1 and feature types that are required for 1:10 000 scale mapping Cartosat-2. Out of these, two GCPs are used for the correction could be satisfactorily identified and captured in Cartosat-1 of Cartosat-1 and Cartosat-2 images, and the rest of the and Cartosat-2 satellite images. points were used as checkpoints. This ensures the same In some cases, features required for larger than input accuracies for both, so that output accuracies will 1:10 000 scale mapping (e.g., roads and woodland indicate the geometric fidelity of individual satellites boundaries at 1:2 500 scale) could also be captured in

1036 September 2010 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING 1031-1040_GS-003.qxd 8/12/10 2:39 PM Page 1037

Figure 10. Ortho-image and DEM generated from Figure 11. Ortho-image and DEM generated from Cartosat-1 (dataset 5): (a) ortho-image, (b) DEM multi-view images of Cartosat-2 (dataset 3): (a) ortho-image, (b) DEM.

the images. Major exceptions to this are transmission lines, walls, fences, and hedges which are generally impossible to distinguish even in imagery with 0.4 m resolution. together, it shows that Cartosat-2 and Cartosat-1 has the Figure 14 show the feature delineation from Cartosat-1 and potential as a data source for 1:7 000 and 1:10 000 scale Cartosat-2 at 1:5 000 scale. As we can see in Figure 14b, mapping, respectively, at the current specification. The sharpness of the features is poor for Cartosat-1 at 1:5 000 advantage of Cartosat-1 satellite over Cartosat-2 is the scale. At the same time, the resolution improvement of stability. The satellite is not changing its view while Cartosat-2 is not making it appropriate for better than imaging, and the image is not acquired in very high 1:7 000 scale mapping from the geometric accuracy point oblique look angles, which is an important criteria for of view. The national mapping accuracy standards in Great mapping. Britain indicate RMSE of 1.1 m for 1:2 500 scale data and Over a built-up area of 2 km ϫ 3 km, a base map was 3.4 m for 1:10 000 scale data. Considering the results of generated at 1:10 000 scale from Cartosat-1 and Cartosat-2, feature capture capability and the geometric accuracy and is shown in Figure 15.

PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING September 2010 1037 1031-1040_GS-003.qxd 8/12/10 2:39 PM Page 1038

Figure 12. Comparison of errors over same check- points with same GCPs used for sensor orientation of Cartosat-1 and Cartosat-2. Figure 13. Portions from ortho-images of dataset 3 (at original resolution): (a) Cartosat-2 and (b) Cartosat-1.

Major road networks and buildings could be captured from both satellite images. The level of features in Cartosat-1 imaging. This ensures that no scale variation will occur at and Cartosat-2 are comparable at 1:10 000 scale. Thus, the different parts of the imagery and will give uniform accuracy generated base map clearly demonstrates the mapping over a single image as well as images taken from different potential of Cartosat imagery. orbits. It is better to use stable satellites than agile ones for mapping applications. • The number of GCP requirement is less for Cartosat-1. One Conclusions conjugate point from Fore and Aft images over a scene gives In this study, we have evaluated the potential of Cartosat-1 an accuracy of about 3 to 4 m. A minimum of two GCPs are and Cartosat-2 satellites for large scale mapping. Following required for correction of Cartosat-2 to compensate for the agility of the satellite. are the major conclusions. • The majority of features seen in Cartosat-2 Images are • Cartosat-1 and Cartosat-2 show that the use of the along delineable in rural and mixed areas in Cartosat-1 images track stereo sensors is very promising for DEM generation. with the desired accuracy. • The accuracy of the sensor model and the high correlation coefficient of the image matching are the two principal factors of getting the expected DEM accuracy. • Cartosat-1 DEM, geometric accuracy, and capability for topo- References graphic feature capture are satisfactory for making 1:10 000 scale Baltsavias, E.P., M. Pateraki, and L. Zhang, 2001. Radiometric and maps. Geometric accuracy and feature detection of Cartosat-2 geometric evaluation of IKONOS Geo images and their use for indicate that it is capable of making 1:7 000 scale maps. 3D building modeling, Proceedings of the Joint ISPRS • The Cartosat-1 satellite is more stable compared to Cartosat-2 Workshop on High Resolution Mapping from Space 2001, because it is not continuously changing the view while 19–21. September, Hannover, Germany, unpaginated CD-ROM.

1038 September 2010 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING 1031-1040_GS-003.qxd 8/12/10 2:39 PM Page 1039

Archives of Photogrammetry and Remote Sensing, 35(B1): 421-232. Radhadevi, P.V., and R. Ramachandran, 1994. Orbit attitude modelling of SPOT imagery with a single ground control point, The Photogrammetric Record, 14(84):973–982. Radhadevi, P.V., S.S. Solanki, R. Ramachandran, R. Krishnan, 2008. Pre-processing consideration of IRS-P6 Liss-4 Imagery, International Journal of Applied Earth Observation and Geoinformation, Special issue of Resourcesat, 10:133–139. Srivastava, P.K., T.P. Srinivasan, A. Gupta, S. Singh, J.S. Nain, Amitabh, S. Prakash, B. Kartikeyan, and B. Gopala Krishna, 2008. Advanced studies in strip processing of Cartosat-1 data, The Photogrammetric Record, 23(123):290–304. Topan, H, G. Büyüksalih, and K. Jacobsen, 2004. Comparison of information contents of high resolution space images, Interna- tional Archives of Photogrammetry, Remote Sensing and Spatial Information Sciences, 35(B4):583–588. Toutin, Th., 2004. Comparison of stereo-extracted DTM from different high-resolution sensors: SPOT-5, EROS-A, IKONOS-II, and QuickBird, IEEE Transactions on Geoscience and Remote Sensing, 42(10):2121–2129.

Figure 14. Feature delineation at 1:5 000 scale: (a) Cartosat-2 and (b) Cartosat-1.

Fraser, C., E.P. Baltsavias, and A. Gruen, 2002. Processing of IKONOS imagery for sub-meter 3D positioning and building extraction, ISPRS Journal of Photogrammetry and Remote Sensing, 56(3):177–194. Grodecki, J., and G. Dial, 2003. Block adjustment of high- resolution satellite images described by rational polynomials, Photogrammetric Engineering & Remote Sensing, 69(1):59–68. Jacobsen, K., 2003. Geometric potential of IKONOS and QuickBird images, Proceedings of Photogrammetric Week 2003 (D. Fritsch, editor), pp. 101–110. Jacobsen, K., 2004. DTM generation by SPOT5 HRS, International Archives of Photogrammetry Remote Sensing and Spatial Information Sciences, 35(B1):439–444. Krishnaswamy, M., 2002. Sensors and platforms for high resolution imaging for large scale mapping applications -Indian scenario, Indian Cartographer, DAPI-01, URL: http://www.incaindia.org/ technicalpapers/02_DAPI01.pdf (last date accessed: 24 June 2010). Poli, D., L. Zhang, A. Gruen, 2004. SPOT-5/HRS stereo image orientation and automatic DSM generation, International

PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING September 2010 1039 1031-1040_GS-003.qxd 8/12/10 2:39 PM Page 1040

Figure 15. Portion of map generated from Cartosat-1 (dotted line) and Cartosat-2 (continuous line) ortho-images overlaid (scale 1:10 000).

1040 September 2010 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING