STEREO EVALUATION OF ALOS/PRISM DATA ON ESA-AO TEST SITES FIRST DLR RESULTS Authors: Mathias Schneider, Manfred Lehner, Rupert Müller, Peter Reinartz Remote Sensing Technology Institute German Aerospace Center ALOS Symposium Rhodos Mathias Schneider November 4, 2008
Outline ALOS/PRISM characteristics Test Data Processing Interior Orientation Exterior Orientation RPC Generation Stereo Processing Forward, Nadir and Backward Coregistration DEM Generation Conclusion 2
The PRISM instrument Japanese satellite ALOS was launched in 2006 3 Instruments onboard: PALSAR, AVNIR and PRISM PRISM consists of 3 panchromatic spectrometers PRISM characteristics: Resolution: 2.5 m Swath width: 35 km (triplet mode) Pointing angle: -1.5 to 1.5 degree Stereo angle: +/- 23.8 degree (F/B) Number of CCDs: 6 (N)/8 (F/B) Pixel per CCD array: 4992 (N)/4928 (F/B) An image is acquired using a subset of 4 consecutive CCD-arrays (triplet mode) From: Tadono et al., 2004 3
Data Different Processing Levels: L1A: No correction L1B1: Images are radiometrically corrected L1B2: Images are radiometrically and geometrically corrected Images are composed by one image file per CCD with an overlap of 32 pixels First step after extraction of image data: Merge image files to one image Images are jpeg compressed Compression artifacts are visible Images are filtered with a Gaussian 3x3 filter Ancillary data is given in leader- and supplemental-file 4
Ancillary Data Ancillary data is given in CEOS-format, partly in ASCII and partly in binary code Extracted data blocks: Precision orbit (Position in Earth Centered Rotated Coordinates, ECR) Precision attitude (Quaternions) Coordinate conversion matrices Geometric parameters (used to generate the interior orientation) 5
Test Data Near Barcelona, Spain 2 Sets of PRISM images acquired in October 2006 (processed at ESA in April 2007) 5 orthophotos DEM (15 m resolution) Near Marseille, France 1 Set of PRISM images acquired in March 2007 (processed at JAXA in October 2007) Coordinates of six GCPs DEM (1 arc sec resolution) 6
Test Data (2) Near Munich, Germany 1 set of PRISM images, acquired in June 2007 (processed at ESA in September 2008) 5 orthophotos DEM (1 arc sec resolution) 7
Processing Direct georeferencing Calculation according to JAXA Algorithm description View vector in CCD coordinate system is calculated (1) CCD-alignment and distortion parameters ( Θ, (k 1/2 ), a, b) are given in supplemental file Linear interpolation (3) View vector is transformed from CCD coordinate system to ECR (ITRF97) (6) Transformation is split up in two parts: Interior orientation Exterior orientation 8
Interior Orientation Interior Orientation has to be given in a table of view vectors for each pixel Coefficients in (9), (10) and (11) are given in the supplemental file (9) accounts for the mount angles of the optics, stereo angles and long period bias time variation (10) accounts for the time, when the satellite is in sunshine (11) accounts for the differences between CCD coordinate system defined in the PRISM sensor model and the reference CCD coordinate system for the pointing alignment parameters 9
Interior Orientation (2) Atmospheric correction has to be applied (12) ΔΘ is a function of Θ, the satellite height and atmospheric parameters For compatibility with existing DLR software, the view vector is normalized the sign of the z-component is changed 10
Exterior Orientation Sensor Positions have to be given in ECR coordinates Hermite Interpolation Orientation angles have to be given in ECR coordinate system Angles are extracted from first part of equation (6) (15) accounts for polar motion, GAST (Greenwich Apparent Sidereal Time) and precession/nutation information and transforms from ECI to ECR Matrix M in (16) is built using the quaternions given in supplemental file Linear interpolation of angles at imaging time Speed of light correction (19) 11
Geolocation DLR software ESTIMATE returns RMS values at GCPs Due to the improvement in the geometric parameters, the location accuracy improved immensely GCPs have to be used to estimate boresight angles for the older datasets GCPs are no longer obligatory for the newest dataset (depending on desired accuracy) Catalonia 25 GCP Marseille 6 GCP Bavaria 33 GCP RMSx [pixel] RMSy [pixel] F 8.187 7.929 N 90.282 15.936 B 38.763 15.988 F 23.405 9.525 N 32.604 2.331 B 4.927 7.168 F 4.787 3.246 N 6.631 3.191 B 2.712 1.071 12
RPC Generation By a modification of the software ORTHO, a three-dimensional grid of control points is generated over the whole image Estimated boresight angles serve as input for ORTHO The RPCs are computed using XDIBIAS RPC generation software To check the RPCs, coordinates of the control points were recalculated using the RPCs and compared to the original coordinates 13
RPC Generation check Oscillation visible in older datasets amplitude of approximately 1 pixel Wavelength approx. 5000 lines, i.e. 1.85 sec or 0.55 Hz DEM-generation might be affected Examination of attitude angles In the newest dataset, the oscillation is no longer visible Either due to the improved sensor model parameters Or the oscillation is caused by something that does not appear permanently (e.g. PALSAR) 14
Attitude angles When plotting the attitude angles, they seem to have a linear behavior Estimation of a second degree Legendre polynomial as trend line and subtraction from the original values Oscillation is clearly visible in older datasets Only very small oscillation with a higher frequency visible in newer dataset 15
Forward, Nadir and Backward Co-Registration Orthoimages are calculated using DLR software ORTHO and 33 GCPs to estimate boresight angles DEM Overlay of Nadir and Backward view indicates very good coregistration Matching between Nadir and Backward orthoimage to display the quality of the coregistration of nadir and backward view Row Column Min -5.14-1.64 Max 4.40 3.08 Mean -1.16 0.57 Std.-dev. 0.64 0.41 16
Image matching for mass tie point generation Hierarchical intensity based matching Reliable tie points for Bundle adjustment Seed points for region growing Automatic 3-ray tie points for GCP Sub-pixel accuracy is achieved with local least squares matching Region growing for densification of tie points based on local least squares matching Blunder reduction includes: Third stereo pair Bi-directional matching 17
DEM Generation Across-track For the Spanish test site, a DEM was calculated and compared to reference DEM RPC-based approach 2 and 3 ray points were used Very good correlation Difference image was calculated Statistics show good accuracy Along-track Overall Chip 1 Min -134.430-81.540 Max 277.740 93.040 Mean 0.936-0.231 Std.-dev. 10.679 3.964 Green: PRISM DEM Blue: Reference DEM 18
Conclusion Extraction of image and ancillary data Direct Georeferencing GCPs are necessary for older datasets, for newer datasets, GCPs are no longer obligatory RPC-based approach reveals oscillation of up to one pixel in the image for older datasets, whereas for the newer dataset, GCPs are no longer obligatory Comparison of PRISM DEM and reference DEM and coregistration of Nadir and Backward image return remarkably good results 19
Thank you for your attention! 20