External quality control of Pléiades orthoimagery

Transcription

1 External quality control of Pléiades orthoimagery Part II: Geometric testing and validation of a Pléiades-1B orthoproduct covering Maussane test site J. Grazzini P. Astrand 2013 Report EUR EN

2 European Commission Joint Research Centre Institute for Environment and Sustainability Contact information P. Astrand Address: Joint Research Centre, Via Enrico Fermi 2749, TP 266, Ispra (VA), Italy Tel.: Fax: This publication is a Reference Report by the Joint Research Centre of the European Commission. Legal Notice Neither the European Commission nor any person acting on behalf of the Commission is responsible for the use which might be made of this publication. Europe Direct is a service to help you find answers to your questions about the European Union Freephone number (*): (*) Certain mobile telephone operators do not allow access to numbers or these calls may be billed. A great deal of additional information on the European Union is available on the Internet. It can be accessed through the Europa server JRC EUR EN ISBN ISSN DOI: /97505 Luxembourg: Publications Office of the European Union, 2013 European Union, 2013 Reproduction is authorised provided the source is acknowledged.

3 i Contents Abstract ii Acknowledgments iii 1 Introduction Input imagery Output orthoimagery generation External geometric quality control Conclusion Annex References List of Illustrations xvi

4 ii Abstract The goal of the current report is to provide with a final statement on the geometric capabilities of Pléiades sensors through the Quality Control (QC) of Pléiades-1B orthocorrected imagery following the recent validation of Pléiades-1A. Namely, the validation process aims at measuring the influence of different factors on the geometric accuracy of Pléiades-1B orthoimagery. For that purpose, a light benchmarking is implemented as: a single far-from-nadir image datasetis evaluated using 2 different software suites only, and with 2 different input GCP configurations, hence meaning that less than 10 products are tested herein. Together with the results of Pléiades-1A validation, it is then asserted that the VHR prime sensor requirement derived from the ASPRS 1: scale map accuracy standards is fulfilled, namely the following condition is met: the planimetric accuracy of the Pléiades orthoimagery, expressed as the RMSE measured on Check Points in both Easting and Northing directions, is below the 2.5 m scale geometric specification. Therefore, Pléiades orthoimagery is formally introduced into the 2013 Control with Remote Sensing Programme.

5 iii Acknowledgments The authors would like to thank F. Ranera (Astrium) for generating the orthoimagery considered in the benchmarking and validation throughout this study. They also acknowledge his numerous comments regarding the outcomes of the benchmarking.

6 4 External quality control of Pléiades orthoimagery 1 Introduction In the framework of the Control with Remote Sensing (CwRS) Programme, the Joint Research Centre (JRC) of the European Commission establishes guidelines for European Member States on the use of Remote Sensing imagery for checking farmers claims for Common Agriculture Policy (CAP). Specifically, the area of claimed parcels is checked using Very High Resolution (VHR) satellite images [1, 2, 3]. Fundamental and critical goals for the purpose of use in CAP checks is the orthorectification of the imagery, a process that enables to georeference the imagery and correct the geometric deformations they undergo during acquisition. The European standard for the satellite orthoimagery requires appropriate quality of the input data (images), as well as the quality assessment of the output (ortho) products. Within this context, Pléiades imagery which offers a 0.5 m Ground Sampling Distance (GSD) that fits CwRS needs should be assessed from a geometric perspective in order to qualify Pléiades 1 as a VHR prime sensor for use in CAP checks i.e. a sensor suitable for measuring parcel areas to the accuracy requested by the CAP regulation. Therefore, the purpose of the current report is to provide with a final statement on the geometric capabilities of Pléiades sensors through the Quality Control (QC) of Pléiades-1B orthocorrected imagery following the prior validation of Pléiades-1A presented in [4]. In practice, data acquired by (early-launched) Pléiades-1A sensor have been already validated through the QC of the orthoimagery in [4]. With the launch of Pléiades-1B sensor, similar experiments need to be undergone. Precisely, the planimetric accuracy of the orthoimagery, derived from ASPRS 1: scale map accuracy standards and expressed as the one-dimensional (i.e. in Easting and Northing directions) Root-Mean-Square Error (denoted RSME 1D ) is requested not to exceed 2.5 m. For this purpose, the processing workflow adopted herein for the quality testing of Pléiades-1B orthoimagery conforms to the standardised methods developed by JRC and derived from either airborne or spaceborne sensors [5]. Under that aspect, it follows the approach of previous HR/VHR orthoimagery validation [6, 7], in particular that of Pléiades-1A [4]. It is based on the benchmarking and testing of few selected Pléiades-1B orthoimages generated within a well-defined context so as to provide a quantitative feedback on the positional accuracy of the sensor. Compared to the benchmarking of Pléiades-1A imagery where a total of 70 different orthoproducts were tested using 3 raw image datasets ranging from close-to- to far-from-nadir, 5 different COTS softwares, and 4 different input GCP configurations a light validation is however implemented for Pléiades-1B, namely: a single far-from-nadir (with 33 incidence angle) image datasetis evaluated using 2 different softwares only (ERDAS and PCI) and with 2 different input GCP configurations, 1 The image provider sells Pléiades imagery with no mention of the version of the specific satellite/sensor 1A or 1B used in acquiring the data. Therefore, the platform has to be considered as a whole in the final statement.

7 2. Input imagery 5 Figure 1: Pan-sharpened Pléiades-1B acquisition over Southern France. Primary image acquired with a viewing angle of 33 over Maussane test site and made available by Astrium for the preparation of the benchmarking. hence meaning that less than 10 products are tested. For further details on the adopted methodology and the underlying motivation for adopting the benchmarking approach, the reader is referred to [4, Section 3], and also [1, 2]. The rest of this report is organised as follows. Next Section describes the input data, i.e. the primary imagery assessed for validation, as well as all other ancillary data used in orthocorrection. Note that Pléiades sensor(s) and its (their) main properties are presented in [4, Section 2]. Section 3 briefly presents the ancillary data used for modeling the orthocorrection and generating the ortho products. The results of the External QC of Pléiades-1B data are presented and discussed in Section 4. Final conclusions and statements regarding Pléiades geometric validation are drawn in Section 5 together with potential improvements for future benchmarking. 2 Input imagery For the various test cases to be elaborated, one single primary image acquired with a large (far-from-nadir) viewing angle over a well-known area is used in input. The test site of Maussane, located in France, had been selected for benchmarking by JRC as it offers sufficient ancillary and reference data (GCPs, DEM) with a validated quality [8]. Following, one AOI was identified for Pléiades validation in particular, it was already used for Pléiades-1A validation [4]: a square subscene of images acquired over Maussane is defined for testing: the AOI covers an extent of (10 10) km 2 with UL corner at position ( E, N) in EPSG (UTM - zone 310 N - ellipsoid WGS84) reference system 2. with dense coverage by existing CPs datasets, 2 The UL corner s location in Geographic Lat/Lon coordinates is: DMS=( E, N), or equivalently DEG=( E, N).

8 6 External quality control of Pléiades orthoimagery Figure 2: Pléiades-1B pansharpened 33 image over Maussane and its footprint. 33 primary products detailed from Figure 1; corresponding footprint in EPSG reference system: blue framebox, and Maussane AOI selected for geometric benchmarking: red framebox. and was covered by the footprint of the acquisition provided by Astrium (33 off-nadir viewing angle). Namely, a single PSH dataset has been acquired for geometric benchmarking over Maussane AOI on 19/03/13 with angle around 33 (Figure 2). 3 Output orthoimagery generation In order to leverage satellite VHR images for applications such as GIS, it is necessary to orthorectify them. The orthocorrection combines relief effects corrections, georeferencing, and high location accuracy [9] of primary images. Overall, the orthocorrection process may be subdivided into two steps [10]: the orientation and the following orthorectification. In the former, an image is georeferenced in a defined reference frame and an analytical relationship between the image and Earth coordinates is established; at this step the distortions due to the acquisition system and to the atmospheric refraction are corrected. In the latter the distortion due to terrain morphology is corrected by means of a suitable Digital Elevation (DEM) or Surface (DSM) Model, starting from the oriented image [11], and using additional location information provided by Ground Control Points (GCPs) covering the AOI. For an image to be accepted (validated), the technical specifications on the auxiliary data (GCPs and DEM/DSM) enounced in JRCguidelines are strictly followed [1]. In this study, a high-resolution/high-precision raster DEM with ellipsoidal heights is used, with the following characteristics: spatial grid of 2 2 m,

9 3. Output orthoimagery generation 7 (a) All 4 input GCPs used in orthocorrection are identified in aerial images of the ADS40 dataset (PAN at 0.5 m resolution) [12], and located in Pléiades-1B images through visual matching [13]. GCPs # ID (b) Left: GCPs selection in primary raw images for the generation of the ortho-products. Right: GCPs (red star ) spatial configuration: 3 and 4 GCP(s) selected. Figure 3: GCPs selection: ground and image identification; configuration. vertical accuracy RMSE Z 0.6 m. as it meets the requirements of JRC guidelines for orthocorrection enounced as [5, Section 8.3]: DEM grid spacing [should be] 5 to 20 times that of the orthoproduct pixel size, depending on the terrain flatness, and DEM height accuracy [should be] 2 planimetric required RSME 1D. The original DEM was produced from digital airborne stereo image pairs (Leica Geosystems) of GSD of 50 cm in the frame of ADS40 project [8]. This DEM is identical to the one used in Pléiades-1A benchmarking, hence the reader is referred to [4] for further details. Control Points (simply denoted CPs) serve for the orthocorrection of the images and

10 8 External quality control of Pléiades orthoimagery PCI ERDAS GCPs East North 2D # [m] [m] [m] Table 1: Results of RMSE 1D measurements over Pléiades-1B 33 image. The RMSEs in Easting, Northing and combined 2D directions are presented alltogether for both PCI and ERDAS softwares. The East (resp., North and 2D) columns store the RMSE 1D [East] (resp., RMSE 1D [North] and RMSE 2D ) errors expressed in meters. the geometric quality validation of the derived orthoimages, provided the fulfilment of the accuracy requirements of JRC guidelines [5, Section 7.1]: GCPs [and ICPs] should be at least 3 times (5 times recommended) more precise than the target specification for the ortho. Likewise the validation of Pléiades-1A data, no new field campaign was operated for collecting GCPs. Instead, GCPs were retrieved from already existing databases of Control Points [14, 8]. In practice, 4 GCPs are selected over the 33 PSH product, namely [8]: 1 GCP is taken from vexcel project database, 3 GCPs are taken from the multipurpose database, and used in two different spatial configurations for benchmarking: see Figure 3. See also Table 2 in Annex 5. The ortho-guideline requirements are met considering the properties of the CPs datasets, as for the positional accuracy: RSME 1D 0.5 m: and the vertical accuracy: RMSE Z 1 m. The reader is referred to [14, 12] for further information regarding the considered Control Points databases. Finally, the definition of an orthorectification model is also required. Such models are usually classified in two categories: physically based models, which take into account several aspects influencing the acquisition procedure, and black-box models which are independent of the sensor s characteristics [11, 4]. Among the second category, the popular Rational Function Method is adopted for the orthorectification of Pléiades-1B VHR images. This method uses the Rational Polynomial Coefficients (RPC) provided with the primary products. In practice, the orthocorrection of Pléiades data was operated by the image provider Astrium using ERDAS and Envi softwares only. 4 External geometric quality control The external quality control of the ortho-rectified product s accuracy is done by measuring the misregistration a given known locations using one parameter: the maximum permissible planimetric error RMSE 1D. Indeed, the so-called Hold-Out-Validation method [7] requires only to check the geometric accuracy on a set of Independent Check Points (ICPs):

11 4. External geometric quality control 9 Figure 4: Representation of RMSE1D errors. From left to right: the planimetric RMSEs in Easting and Northing directions are displayed alltogether (left); 2D representations of the errors estimated over the ortho outputs produced using 3 (middle) and 4 (right) GCPs resp.. On those images, the error is represented over the ICPs as as red arrow ( ) that indicates the shift induced by the orthocorrection: its width represents the RMSE estimated locally at the considered ICP (the larger the arrow, the bigger the RMSE), while its width and orientation depend on the horizontal errors RMSE1D [North] and RMSE1D [East] at that same ICP. Results are presented for both softwares used in the orthorectification process: PCI (top) and ERDAS (bottom). that were not included in the orthocorrection model definition, whose ground coordinates are (known and) derived from other (possibly more accurate) source, whose image coordinates are (identified and) used as reference, that remains unchanged for all tested ortho products. In practice, thirty six ICPs have been retrieved from the same CPs datasets [14, 8] as the one considered for defining the GCPs, and are used for the geometric evaluation of Ple iades-1b orthoimagery: the accuracy is evaluated as the RMSE1D of the residuals (see [4, Section 6.2]) between the orthoimagery derived coordinates of this set of points and their true ground coordinates. A quick look at the results presented in Table 1 and Figure 4 show that: there is an important shift in South-East direction (overall directions of arrows representing the error),

12 10 External quality control of Pléiades orthoimagery errors along Easting direction higher than those along Northing, which seems in contradiction with the general trend observed on Pléiades-1A. In addition, it is observed that using PCI (see Figure 4), there is a rather significant overall improvement in the orthocorrection accuracy (RMSE 2D ) when using 4 GCPs instead of 3 only (see Table 1). In general, errors are higher in montainous areas (central and Northern parts of the image, see Figure 4). The error over ICPs located in the South-Eastern area part of the image is larger for the 3-GCPs based product as this area is not covered with any GCP. Instead, using ERDAS (see also Figure 4), the improvement observed when using 4 GCPs is not so significative. Likewise the PCI products, higher errors are observed in the montainous areas. However, the use of a 4th GCP does not affect the orthocorrection accuracy in the South-Eastern area that was not covered with 3 points only. In general, improved results were found using PCI platform. In all cases, the planimetric accuracy of the Pléiades-1B orthoimagery in both Easting and Northing directions is below the 2.5 m scale geometric specification. 5 Conclusion The External Quality Control assessing Pléiades orthoimagery from a geometric perspective has been performed in accordance with the standard methodology and guideline enounced by JRC. The validation test protocol, as well as the statistical analysis of the data, adopted for the qualification of both Pléiades-1A and Pléiades-1B sensors is identical, with however a difference in terms of total amount of tested data. A benchmarking aiming at both evaluating the geometric (positional) accuracy of Pléiades orthoimagery, and measuring the influence of different factors (viewing angle, number of GCPs, orthorectification model) on the accuracy has been operated separately, as: for Pléiades-1A, 3 raw image datasets ranging from close-to-nadir from far-from-nadir (namely, with 4, 22, and 30 incidence angles) have been tested using 4 different COTS softwares (namely, ERDAS, Envi, PCI and Keystone) plus one in-house software (PF), and 4 different input configurations, hence leading to the testing of around 70 different products; for Pléiades-1B (launched after Pléiades-1A), only one far-from-nadir (with 33 incidence angle) image dataset has been evaluated using 2 different softwares only (ERDAS and PCI) and with 2 different input configurations, hence meaning that less than 10 products have been tested. Following the testing of Pléiades-1B geometric quality presented in this report, Pléiades-1B sensor is validated from the geometric perspective. Together with the results of Pléiades-1A validation already presented in [4], it is asserted that the VHR prime sensor requirement derived from the ASPRS 1: scale map accuracy standards is fulfilled, namely the following condition is met: the planimetric accuracy of the Pléiades orthoimagery, expressed as the RMSE measured on Check Points in both Easting and Northing directions, is below the 2.5 m scale geometric specification.

13 5. Conclusion 11 In complement to the studies presented in this report and [4], it is believed that some further testing/benchmarking should be performed: bundle (PAN, MSP) image products should be tested vs. pansharpened (PSH) ones, more detailed tests on the minimum number (1, 2 or 3) of GCPs necessary for adequate orthorectification should be operated, an estimate of the extreme viewing angle acceptable for orthocorrection should be derived, new sites with different terrain types (high topography, heterogeneous terrain) should be tested, further tests on RPC, and rigorous models should be operated, a full scene testing should be scheduled, as to in-depth validate the planimetric positional accuracy of the produced orthoimage. This validation should also be the subject of additional testing procedure on image data acquired during the agriculture season. Indeed, the overall radiometric quality and image content of the orthoimagery products should also be tested and validated (see [4, Annex A.3]). The buffer width has not been formally tested but is assumed to be satisfied. Considering the absence of parcel area measurement validation results, a value equal to 1.5 times the pixel size should be used as a reasonable approximation: the tolerance will be 0.75 m times parcel perimeter length.

14 12 External quality control of Ple iades orthoimagery Annex: Description of ancillary GCPs Tables 2 and 3 display some of the metadata describing and helping at (visually) extracting the GCPs used in orthorectification (see Section 3): ground pictures and image chips, ground positions, plus the derived image locations. # ID source multipurpose Vexel multipurpose multipurpose screen shot ground camera shot Table 2: GCPs selection over Mausanne site. GCPs from 2 different datasets were selected and positioned on the primary imagery based on the available visual information (ground camera shots and image screenshots).

15 5. Conclusion 13 ground position height image location [m] [m] [pixels] # ID North East ellips. ortho. X Y N/A N/A N/A Table 3: Ground position, height and image location of selected GCPs. In-situ measured GPS (North, East) coordinates in EPSG reference system and respective heights. (column,row) image coordinates (X, Y ) in 33 image are identified by a human operator (1 digit precision, i.e. a tenth of a pixel).

16 14 External quality control of Pléiades orthoimagery References [1] S. Kay, P. Spruyt, and K. Alexandrou, Geometric quality assessment of orthorectified VHR space image data, Photogrammetry Enginnering and Remote Sensing, vol. 69, no. 5, pp , [2] J. Chmiel, S. Kay, and P. Spruyt, Orthorectification and geometric quality assessment of very high spatial resolution satellite imagery for Common Agricultural Policy purposes, in Proc. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 2004, vol. 35. [3] M.A. Aguilar, F. Agüera, F.J. Aguilar, and F. Carvajal, Geometric accuracy assessment of the orthorectification process from very high resolution satellite imagery for Common Agricultural Policy purposes, International Journal of Remote Sensing, vol. 29, no. 24, pp , [4] J. Grazzini, S. Lemajic, and P.J. Åstrand, External quality control of Pléiades orthoimagery Part I: Geometric benchmarking and validation of Pléiades-1A orthorectified data acquired over Maussane test site, Tech. Rep , JRC IES, [5] D. Kapnias, P. Milenov, and S. Kay, Guidelines for best practice and quality checking of ortho imagery, Tech. Rep , JRC IPSC, 2008, available at ec.europa.eu/mars/content/download/1231/7140/file/orthoguidelines_v3_ final.pdf; wiki available at php/guidelines_for_best_practice_and_quality_checking_of_ortho_imagery. [6] J. Nowak Da Costa and A. Walczyńska, Geometric quality testing of the WorldView-2 image data acquired over the JRC Maussane test site using ERDAS LPS, PCI Geomatics and Keystone digital photogrammetry software packages Initial findings with annex, Tech. Rep , JRC IPSC, 2011, available at bitstream/ /22790/1/jrc60424_lb-nb-24525_en-c_print_ver.pdf. [7] P.J. Åstrand, M. Bongiorni, M. Crespi, F. Fratarcangeli, J. Nowak Da Costa, F. Pieralice, and A. Walczynska, The potential of WorldView-2 for ortho-image production within the Control with Remote Sensing Programme of the European Commission, International Journal of Applied Earth Observation and Geoinformation, vol. 19, pp , [8] J. Nowak Da Costa and P.A. Tokarczyk, Maussane test site auxiliary data: existing datasets of the Ground Control Points, Tech. Rep , JRC IPSC, [9] W. Linder, Digital Photogrammetry A Practical Course, Springer, 2009, 3 rd edition.

17 References 15 [10] T. Toutin, Geometric processing of remote sensing images: models, algorithms and methods, International Journal of Remote Sensing, vol. 10, pp , [11] T. Toutin, R. Chénier, and Y. Carbonneau, 3D models for high resolution images: Examples with QuickBird, IKONOS and EROS, in Proc. International Society for Photogrammetry and Remote Sensing Congress, 2003, pp [12] J. Nowak Da Costa and P.A. Tokarczyk, The GCP dataset based on the Leica Geosystems ADS40 digital airborne camera orthoimagery, Tech. Rep , JRC IPSC, [13] J.A. Gutierrez and B.S.R. Armstrong, Precision Landmark Location for Machine Vision and Photogrammetry Finding and Achieving the Maximum Possible Accuracy, Springer-Verlag, London (UK), [14] C. Lucau and J. Nowak Da Costa, Maussane GPS field campaign: methodology and results, Tech. Rep , JRC IPSC, 2009, available at /1/pubsy_jrc56280_fmp11259_sci-tech_report_cl_jn_mauss pdf.

18 xvi List of Illustrations Figures 1 Large pan-sharpened Pléiades-1B acquisition over Southern France Pan-sharpened Pléiades-1B 33 image over Maussane and its footprint GCPs selection, ground identification and image configurations a Ground identification of GCPs in aerial images b Spatial configurations of GCPs Representation of RMSE 1D errors Tables 1 Results of RMSE 1D measurements over Pléiades-1B 33 image GCPs selection over Mausanne site Ground position, height and image location of selected GCPs

19 European Commission EUR Joint Research Centre Institute for Environment and Sustainability Title: External quality control of Pleiades orthoimagery Part II: Geometric testing and validation of a Pléiades-1B orthoproduct covering Maussane test site Author(s): J. Grazzini and P. Astrand Luxembourg: Publications Office of the European Union pp x 29.7 cm EUR Scientific and Technical Research series ISSN ISBN DOI: /97505 Abstract The goal of the current report is to provide with a final statement on the geometric capabilities of Pléiades sensors through the Quality Control (QC) of Pléiades-1B orthocorrected imagery following the recent validation of Pléiades-1A. Namely, the validation process aims at measuring the influence of different factors on the geometric accuracy of Pléiades-1B orthoimagery. For that purpose, a light benchmarking is implemented as: a single far-from-nadir image datasetis evaluated using 2 different software suites only, and with 2 different input GCP configurations, hence meaning that less than 10 products are tested herein. Together with the results of Pléiades-1A validation, it is then asserted that the VHR prime sensor requirement derived from the ASPRS 1: scale map accuracy standards is fulfilled, namely the following condition is met: the planimetric accuracy of the Pléiades orthoimagery, expressed as the RMSE measured on Check Points in both Easting and Northing directions, is below the 2.5m scale geometric specification. Therefore, Pléiades orthoimagery is formally introduced into the 2013 Control with Remote Sensing Programme.

20 z LB-NA EN-N As the Commission s in-house science service, the Joint Research Centre s mission is to provide EU policies with independent, evidence-based scientific and technical support throughout the whole policy cycle. Working in close cooperation with policy Directorates-General, the JRC addresses key societal challenges while stimulating innovation through developing new standards, methods and tools, and sharing and transferring its know-how to the Member States and international community. Key policy areas include: environment and climate change; energy and transport; agriculture and food security; health and consumer protection; information society and digital agenda; safety and security including nuclear; all supported through a cross-cutting and multidisciplinary approach.