Presentation is loading. Please wait.

Presentation is loading. Please wait.

Alos optical data processing End of Stage Presentation Trainee: ARNAL Etienne Tutor: Mr COLIN Olivier

Similar presentations


Presentation on theme: "Alos optical data processing End of Stage Presentation Trainee: ARNAL Etienne Tutor: Mr COLIN Olivier"— Presentation transcript:

1 Alos optical data processing End of Stage Presentation Trainee: ARNAL Etienne Tutor: Mr COLIN Olivier

2 18/05/2014 ESRIN - ARNAL Etienne 2 Presentation Outline Alos Product Descriptor (PRISM and AVNIR2) Alos Image Processing (PRISM and AVNIR2) Digital Elevation Model 3D results

3 18/05/2014 ESRIN - ARNAL Etienne 3 Alos Product Descriptor (1/5) ImageDEM Realisation of a cqfd dictionary containing the fields of the Product Format Description of Prism and Avnir2. Allows to find quickly all fields of level: 1A = raw data extracted from the Level 0, 1B1 = with radiometric corrections, 1B2R = geo-referenced, 1B2G = geo-coded. 3D results Data

4 18/05/2014 ESRIN - ARNAL Etienne 4 Alos Product Descriptor (2/5) CQFD: binary data reader. VOL LED IMG TRL SUP Tie Points in the data FILE ; |_ VOL ; | |_ VolumeDescriptor ; | |_ FilePointer ; (3 to 9) | |_TextRecord ; |_ LED ; | |_ CommonFileDescriptor ; | |_ FileDescriptor ; | |_ SceneHeader ; | |_ Ancillary1 ; | |_ Ancillary2 ; | |_ Ancillary3 ; DataImageDEM 3D results

5 18/05/2014 ESRIN - ARNAL Etienne 5 Alos Product Descriptor (3/5) CQFD: binary data reader. VOL LED IMG TRL SUP Tie Points in the data |_ IMG ; | |_ IMG1 ; (1 to 6) | | |_ CommonFileDescriptor ; | | |_ FileDescriptor ; | | |_ Line 1 ; | | |_ … | | |_ Line n ; |_ TRL ; | |_ CommonFileDescriptor ; | |_ FileDescriptor ; | |_ Trailer ; |_ SUP ; | |_ CommonFileDescriptor ; | |_ FileDescriptor ; | |_ Ancillary ; DataImageDEM 3D results

6 18/05/2014 ESRIN - ARNAL Etienne 6 Alos Product Descriptor (4/5) CQFD: binary data reader. VOL LED IMG TRL SUP Tie Points in the data |_ TiePoints ; | |_ Line 1 ; | | |_ Lat ; | | |_ Long ; | |_ … | |_ Line n ; (I,J) are pixel and line of corrected image (φ,λ) are degrees of latitude and longitude DataImageDEM 3D results

7 18/05/2014 ESRIN - ARNAL Etienne 7 Alos Product Descriptor (5/5) Thanks to this CQFD dictionary, everybody can find a precise field in every Prism and Avnir2 data. The next step is to create image. DataImageDEM 3D results

8 18/05/2014 ESRIN - ARNAL Etienne 8 Alos Image Processing (1/5) 3 various image data structure: Several CCD (Prism 1A, 1B1) From 3 to 6 CCD CCD = ~4992 pixels * lines Overlap = 32 pixels One CCD (Prism 1B2R, 1B2G) CCD = ~15000 pixels * lines 4 Spectral Bands (Avnir2) Band = ~7150 pixels * 8000 lines DataDEM 3D results Image

9 18/05/2014 ESRIN - ARNAL Etienne 9 Alos Image Processing (2/5) Imager: data reader for image generation For each pixel: –elaborates from the band values (0 to 255) applying sun angle correction and stretching to enhance image contrast, –extracts Tie Point (λ,φ) grid. It creates: one image file (including all CCD/Bands), one GeoTiff file (geolocalisation). DataImageDEM 3D results

10 18/05/2014 ESRIN - ARNAL Etienne 10 Prism Nadir 1B2R DataImageDEM 3D results Shizuoka, Japan on February 14, 2006 Alos Image Processing (3/5)

11 18/05/2014 ESRIN - ARNAL Etienne 11 Alos Image Processing (4/5) Avnir2 DataImageDEM 3D results Tanegashima Is., Japan on February 17, 2006

12 18/05/2014 ESRIN - ARNAL Etienne 12 Alos Image Processing (5/5) encountered problems Avnir2, ±5 pixels between each columns. Prism and Avnir2, geolocalisation error. Prism, many artifacts. Prism and Avnir 1B2(R/G), nothing about time in the data. DataImageDEM 3D results NadirForward

13 18/05/2014 ESRIN - ARNAL Etienne 13 Digital Elevation Model (1/10) Stereoscopy 3D, imager process. For each pixel [I,J], Find the value of pixel Nadir [I,J] and [λ,φ]=f N (I,J) Find the value of pixel Backward f B -1 [λ,φ] Find the value of pixel Forward f F -1 [λ,φ] Output: one RGB image with R=Nadir, G=Forward and B=Backward. DataImage 3D results DEM

14 18/05/2014 ESRIN - ARNAL Etienne 14 Digital Elevation Model (2/10) DEM, bases Height is proportional to the along track displacement. DataImageDEM 3D results Nadir Forward Satellite track

15 18/05/2014 ESRIN - ARNAL Etienne 15 Digital Elevation Model (3/10) DEM, base equation is a multiple of the ground resolution DataImageDEM 3D results

16 18/05/2014 ESRIN - ARNAL Etienne 16 Digital Elevation Model (4/10) DEM Algorithm. Image adjustments DataImageDEM 3D results ForwardNadirBackward Mean = 35 STD = Mean = 43 STD = Mean = 57 STD =

17 18/05/2014 ESRIN - ARNAL Etienne 17 Digital Elevation Model (5/10) DEM Algorithm. Image adjustments, Generate pyramid image, DataImageDEM 3D results with gradually degraded resolution. Pass the matched points to the higher resolution level for fine matching.

18 18/05/2014 ESRIN - ARNAL Etienne 18 Digital Elevation Model (6/10) DEM Algorithm. Image rectification, Generate pyramid image, Feature matching, DataImageDEM 3D results

19 18/05/2014 ESRIN - ARNAL Etienne 19 Digital Elevation Model (7/10) DEM Algorithm. Image rectification, Generate pyramid image, Feature matching, Matching results, DataImageDEM 3D results 1/16 1/8 1/4 1/2 1/1

20 18/05/2014 ESRIN - ARNAL Etienne 20 Digital Elevation Model (8/10) DEM Algorithm. Image rectification, Generate pyramid image, Feature matching, Matching results, Filtering. DataImageDEM 3D results Median filter, reduce noise (peaks): Mean filter, reduce high frequencies:

21 18/05/2014 ESRIN - ARNAL Etienne 21 Digital Elevation Model (9/10) Integration in eogrid.esrin.esa.int. Modification for Linux (no graphics), Installation of Matlab Component Runtime, Compilation of Matlab code (Stand-alone application). Integrated new DEM application in G-POD in < 1 day DataImageDEM 3D results

22 18/05/2014 ESRIN - ARNAL Etienne 22

23 18/05/2014 ESRIN - ARNAL Etienne 23 Digital Elevation Model (10/10) encountered problems Co-registration is corrected manually due to ALOS processors geo- referencing problem Result of matching requires validation of the correlation distribution not trivial Correlation is impaired by high height gradients in image (e.g. buildings) possible solution to weight the matching window with distance to centre Patterns oriented along-track have low correlation contrast Moving objects (cars, boats, troy horses, etc) Sun glint creates big differences between views Image radiometric quality (e.g. striping, spurious saturation artifacts) Image boundaries DataImageDEM 3D results

24 18/05/2014 ESRIN - ARNAL Etienne 24 3D results Japan, Shizuoka and surroundings, 02/14/2006. AATSR Hurricane Jeanne, 20/09/2004 Stereoscopic views. Digital Elevation models. DataImageDEM 3D results

25 18/05/2014 ESRIN - ARNAL Etienne 25 Potential Enhancements Investigate alternative matching methods for high gradient terrains (e.g. parallax, urban areas) Make adaptive algorithm e.g. apply dynamically different techniques/parameters depending on correlation contrast observed Integrate automatic co-registration of the 3 views with a good geo-referencing. Add possibility to correlate several bands in parallel (e.g. AATSR dual views with 4 channels) Include DEM geo-coding in last stage Optimize algorithm for performance in multi-processing environment (e.g. tiling) Calibration & Validation !

26 18/05/2014 ESRIN - ARNAL Etienne 26 Thanks Olivier Colin, Jerome Louis, Philippe Goryl, Peter Reinartz (DLR), Nityaporn Sirikan, Patrick Grimont, Emmanuel Mathot. Questions ? DataImageDEM 3D results


Download ppt "Alos optical data processing End of Stage Presentation Trainee: ARNAL Etienne Tutor: Mr COLIN Olivier"

Similar presentations


Ads by Google