LPIS Orthoimagery An assessment of the Bing imagery for LPIS purpose Slavko Lemajić Wim Devos, Pavel Milenov GeoCAP Action - MARS Unit - JRC Ispra Tallinn, 24 th November 2011 Outline JRC`s Ortho specifications for LPIS and CwRS Orthoimagery QA and QC Alternative sources of standard reference data: BING Maps Orthoimagery production Quality assesment of Bing Maps Test method Results Comparision with other orthoimagery Conclusions 1
Quality issues QA and QC Specification http://mars.jrc.ec.europa.eu/mars/content/download/1231/7 140/file/Orthoguidelines_v3_final.pdf Guidelines for Best Practice and Quality Checkingof Ortho Imagery http://marswiki.jrc.ec.europa.eu/wikicap/index.php/orthoim age_technical_specifications_for_the_purpose_of_lpis Orthoimage technical specifications for the purpose of LPIS Establish a core set of measures to ensure sufficient image quality for the purposes of LPIS Define the set of metadata necessary for data documentation and overall job tracking Image resolution Data acquisition Data processing Orthoimage Metadata Data Quality Measure Orthoimagery Specification http://marswiki.jrc.ec.europa.eu/wikicap/index.php/orthoimage_technical_specifications_for_the_pur pose_of_lpis Common Specifications DQ Sub elements Conformance Quality Level & Tolerance Limits Notes Expected rate of conforming items Spatial resolution <= 1m Ratio of the final ortho resolution to the GSD is 1:1 for digital sensors, whereas for film cameras should be at least 1.2:1 100% Radiometric resolution =>8 bits/channel 11-12 bits per channel is highly recommended 100% Spectral resolution Color (natural or color infrared) Panchromatic only (satellite or aerial) data is allowed, only if there is no option for color imagery 100% General Image Quality Checking for existence of scratches, dust, threads, hot Lack of defects and artifacts, which could prevent spots, haze, drop lines, shadows, color seams, spilling, N/A (no defects the visual interpretation of the image artifacts, etc allowed) Cloud cover <5-10% Per image and in total. The term "image" is used for the control unit e.g. orthoimage, mosaic (map sheet) 100% Overall clliping <0.5% at each tail Overall clipping of the luminosity histogram 100% Histogram Peak +/-15% of middle value For 8 bit image, the middle value is 128 100% Color balance <2% between min and max value of triplet Not applicable for panchromatic only 100% Noise Contrast Geometric accuracy Mosaicking Mosaicking Compression Signal to Noise Ratio > 12 for each channel The coefficient of variation of the image DN values should be in the range of 10-20% RMSEx <= 2.5 m; RMSEy <= 2.5 m DN valies variation on similar area type not to exceed 10% in average (or 4% between each of the 3 channels) Geometric mismatches along seam lines (d) <3 pixels Lossless (TIF, LZW-TIF) Visually lossless (JPEG2000, ECW, MrSID) at last stage (storage) SNR which is defined as the ratio of the mean DN value to the standard deviation of the DN values 100% Represented as the Standard Deviation of the DN values as a percentage of the available grey levels 100% RMSE is calculated on the base of at least 20 well distributed independent check points (ICP), per image 100% 2
Source of Orthoimagery for LPIS QA CwRS http://mars.jrc.it/mars/content/download/987/6070/file/82 18_zone_selection.pdf Aerial Orthoimagery from national campaigns Orthoimagery from other sources (commercial that meet Specifications) Image acquisition and GCP`s Image acquisition Purpose Airborne sensor (specification, calibration) or spaceborne resolution(spatial, spectral, radiometric) Overlaps (p, q) GCP distribution GCP Field measurement (GPS) Points visible on Images (accuracy, RMSE) 3
Photogrammetric processing and DTM Aerialtriangulation: Input: Imagery GCP`s Adjustment Measures (tie points and GCP`s) Adjustment (RMSE tie points and GCP`s) Aerial oriented images (absolute orientation) DTM Use of existing DTM Check for accuracy Check for completeness Update if necessary Producing the new DTM Specification http://marswiki.jrc.ec.europa.eu/wikic ap/index.php/digital_elevation_model Orthoimagery Orthoimagery: Predecesors AT (oriented images) DTM Orthocorection Proper software Control with DTM Orthomosaic Seamless Radiometric correct http://mars.jrc.ec.europa.eu/mars/content/download/1231/7140/file/ Orthoguidelines_v3_final.pdf 4
An assessment of the Bing imagery for LPIS purpose Outline: BING test over Maussane data preparation and evaluation method First Results geometric accuracy and radiometric accuracy (visual check) Examples QC results BING Maps Data delivered Metadata QC records RGB_N43E004Test_2011_05_28_Q UALITY.xml RGB_N43E004Test_2011_05_28_Q UALITY.txt Data itself RGB ortho images (GSD 30cm) CIR ortho images (GSD 60 cm) Data coverage 5
BING Maps Data delivered CIR RGB Maussane area Spaceborne sensors Airborne sensors LANDSAT 5 (GSD = 28.5 m) Leica ASD40 PAN (GSD = 0.5 m) LANDSAT 7 (GSD = 14.25 m) Vexcel Ultracam colour (GSD = 0.5 m) FORMOSAT-2 MS (GSD = 8 m) Bing utlracamm xd data for Mausanne (GSD = 0.3 m) RGB, 2011 RapidEye (GSD = 6.5 m) Bing utlracamm xd data for Mausanne (GSD = 0.6 m) CIR, 2011 FORMOSAT-2 PAN (GSD = 2 m) UAV (CSD=0.1 and 0.2 m) Maussane campaign September 2011 EROS A PAN (GSD = 1.8 m) Mausanne VW2_L - zenith angle 21.8 degree, 22.01.2010 IKONOS PAN (GSD = 1.0 m) Mausanne VW2_H zenith angle 35.7 degree, 18.04.2010 EROS B (GSD = 0.7 m) Mausanne GE1 L - zenith angle 8.3 degree, 27.01.2009 QUICKBIRD Pansharpened (GSD = 0.6 Mausanne GE1 H - zenith angle 27.6 degree, 27.01.2009 m) WorldView - 1 PAN (GSD = 0.5 m) Bing utlracamm xd data for Mausanne (GSD = 0.3 m) RGB GeoEye - 1 Pansharpened (GSD = 0.5 m) Bing utlracamm xd data for Mausanne (GSD = 0.6 m) CIR GeoEye - 1 Panchromatic (GSD = 0.5 m) UAV for Mausanne (GSD = 0.1 m) RGB, campaign September 2011 EROS B Panchromatic (GSD = 0.7 m) WorldView 2 - Panchromatic (GSD = 0.5 m) WorldView - 2 Pansharpened (GSD = 0.5 m) 6
QUICKBIRD Pansharpened (GSD = 0.6 m) Leica ASD40 PAN (GSD = 0.5 m) 7
Vexcel Ultracam colour (GSD = 0.5 m) Bing utlracamm xd data for Mausanne (GSD = 0.3 m) RGB, 2011 8
BING Maps QC method QC of data delivery QC against existing orthoimagery QC geometric accuracy (against GCP`s) QC radiometric accuracy (visual check) GCP (1) Reference GCP`s and Orthoimagery Maussane GPS measurements for Vexcel project (2005) 25 points (11 suitable for QC) RMSE_X=0.49 m, RMSE_Y=0.50m Geometric Quality Control of an Ultracam Ortho-rectified image (JRC, 2005) (2) Reference GCP`s and Orthoimagery Maussane GPS measurements for ADS40 project (2003) 56 points (10 suitable for QC) RMSE_X<5cm, RMSE_Y=10cm Maussane Test Site Auxiliary Data: Existing Datasets of the Ground Control Points (JRC, 2010) 9
BING Maps Evaluation method (for GCP s) Compare image coordinates with reference dataset image coordinate referent coordinate QC Records 1st dataset GCP s distribution Geometric accuracy: RMSEx 0,46 m RMSEy 1,01 m 10
QC Records 2nd dataset GCP s distribution Geometric accuracy RMSEx 0,25 m RMSEy 1,02 m Examples fit against old orthophoto (Bing CIR) 11
Examples fit against old orthophoto (Bing RGB) Examples fit against old orthophoto (Vexcell RGB) 12
Examples General Image Quality visual check - shadows Shadow 16m Shadow 45m QC Bing assesment DQ Sub elements Spatial resolution Conformance Quality Level & Tolerance Limits <= 1m Results Passed Radiometric resolution =>8 bits/channel Passed Spectral resolution General Image Quality Color (natural or color infrared) Lack of defects and artifacts, which could prevent the visual interpretation of the image Passed Passed with comments Cloud cover <5-10% Overall clliping <0.5% at each tail Histogram Peak +/-15% of middle value Passed Not controled Not controled Color balance <2% between min and max value of triplet Not controled Noise Signal to Noise Ratio > 12 for each channel Not controled Contrast Geometric accuracy Mosaicking Mosaicking Compression The coefficient of variation of the image DN values should be in the range of 10-20% RMSEx <= 2.5 m; RMSEy <= 2.5 m DN valies variation on similar area type not to exceed 10% in average (or 4% between each of the 3 channels) Geometric mismatches along seam lines (d) <3 pixels Lossless (TIF, LZW-TIF) Visually lossless (JPEG2000, ECW, MrSID) at last stage (storage) Not controled Passed Not controled Passed Passed 13
Conclusion Bing Data delivery More metadata needed (image acquisition, DTM used) QC records should follow the Specification (metadata) Geometric accuracy Data is compliant for LPIS update and LPIS QA Color and CIR Orthoimages fit correctly Bing maps fit correct against old orthoimages Radiometric accuracy (visual check) Generaly images are acceptable but some issues are observed long shadows Oblique acquisition in mountain area The complete test should be performed (all QE, analysis of image content (CIR, Color and VHR)) Conclusion Common specification QE (all must passed) Importance of quality of orthoimagery For creation and updating of LPIS For ETS Level of Quality should not be reduced 14
Thank you for your attention! slavko.lemajic@jrc.ec.europa.eu wim.devos@jrc.ec.europa.eu pavel.milenov@jrc.ec.europa.eu 15