Test Ranges for Metric Calibration and Validation of Satellite Imaging Systems Gene Dial Jacek Grodecki
© 2003 Space Imaging, LLC. Confidential and proprietary. Topics IKONOS Introduction Importance of Clear Skies Preflight test ranges Requirements Validation Post Launch Test Ranges Recalibration Accuracy Test & Monitoring Ground Control Points Good & Bad Image Display Effects High Quality Summary & Conclusions
© 2003 Space Imaging, LLC. Confidential and proprietary. IKONOS Satellite Use Simultaneous Pan and Multispectral images Panchromatic pixels: 0.82 m at nadir Multispectral pixels: 3.28 m at nadir (Blue, Green, Red, NIR) Linear detector arrays scan in any direction Swath width: 10.8 km at nadir Image Specifications Use 680-km near-polar, sun-synchronous orbit. 3-day revist with elevation > 60° 25 m CE90 accuracy without GCP 2 m CE90 accuracy with GCP Same pass stereo strips up to 400 km long 25 m CE90 horizontal, accuracy 22 m LE90 vertical accuracy Metric Specifications
© 2003 Space Imaging, LLC. Confidential and proprietary. Select Low-Cloud Areas for Test Ranges Gene Dial, SIE, 4/5/97
© 2003 Space Imaging, LLC. Confidential and proprietary. Northern Test Range Purpose: Verify Requirements Size: 20 km NS x 30 km EW GCP: 140 cultural features Accuracy: 0.5 m CE90, 0.9 m LE90 Source: Large-scale aerial photography Imagery: “San Diego Seven” Stereo triplet on the West Stereo pairs center and East Supplied to SW developers for test Results: IKONOS meets requirements. Without Ground Control – 25 m CE90 horizontal – 22 m LE90 vertical With Ground Control – 2 m CE90 horizontal – 3 m LE90 vertical
© 2003 Space Imaging, LLC. Confidential and proprietary. Southern Test Range 200 GCP in 100 km x 100 km area Differential GPS Survey NS line for exterior orientation EW line for interior orientation Many points difficult to identify Chip at lower right shows 1 m by 2 m black rectangle on 1 m diameter white water pipe
© 2003 Space Imaging, LLC. Confidential and proprietary. Interior & Exterior Orientation
© 2003 Space Imaging, LLC. Confidential and proprietary. Case Study: 6 Stereo Pair, 60 GCP Miles N EW S Processed with one GCP. Remaining GCP test accuracy.
© 2003 Space Imaging, LLC. Confidential and proprietary. Original Field Angle Map, 1 GCP CE90 = 8.6 m Project Area 6 Stereo Pair Wide Precision GCP Residual Error Pattern All residuals point to GCP Suggestive of scale error About 2 m per strip Corrective Action Denver test range for FAM calibration Interior orientation adjusted Scale Error Removed
© 2003 Space Imaging, LLC. Confidential and proprietary. Calibration of Interior Orientation
© 2003 Space Imaging, LLC. Confidential and proprietary. After FAM Recalibration of Nov 19, CE90 = 1.8 m
© 2003 Space Imaging, LLC. Confidential and proprietary. Exterior Orientation Repeated acquisitions of MTF target Target surveyed by differential GPS Median error measured Interlock angles adjusted to zero median error. Initialization
© 2003 Space Imaging, LLC. Confidential and proprietary. Repeated Acquisitions to Verify Interlock Angles
© 2003 Space Imaging, LLC. Confidential and proprietary. Ad-hoc Tests of System Accuracy Procedure Acquire GCP for Precision orders Process Imagery without GCP Evaluate Errors Example 49 check points 24 stereo pairs Reference: Dial, IKONOS Stereo Accuracy, ASPRS GCP: Cultural Features Range-only differential 1 m CE90, 2 m LE90. Result: IKONOS Stereo Accuracy 6 m CE90 Horizontal 10 m LE90 Vertical
© 2003 Space Imaging, LLC. Confidential and proprietary. Repeatability Results of 5 trials with 1 operator Range = pix Sigma = 0.16 pix Operator variation could be significant
© 2003 Space Imaging, LLC. Confidential and proprietary. Effect of Resampling and Contrast on Corner Position Change from NN to CC resampling moves apparent corner position by 0.25 pixels
© 2003 Space Imaging, LLC. Confidential and proprietary. Contrast: Low vs. High
© 2003 Space Imaging, LLC. Confidential and proprietary. Brightness: Light & Dark
© 2003 Space Imaging, LLC. Confidential and proprietary. Effect of Resampling, Brightness, & Contrast Resampling effect is in the noise level. Contrast changes had no effect on perceived position. Brightness changes had a significant effect Bright images – Lower contrast to white square – Increase contrast to black square – Operator picks further away from black corner Dark images have converse effects Ideal GCP are centerline, not corner features.
© 2003 Space Imaging, LLC. Confidential and proprietary. Examples of Good Ground Control Good Ground Control Points Intersection of Cross-Walk and Stop Lines Sidewalk intersection Home plate at baseball field Intersection of Service Line and Center Line of a Tennis Court – Intersection of Side Lines with Base Line is not good.
© 2003 Space Imaging, LLC. Confidential and proprietary. Examples of Poor Ground Control Examples of Poor GCP Locations Dirt Road Intersection Acute road intersection Walled compounds
© 2003 Space Imaging, LLC. Confidential and proprietary. GCP Panels for IKONOS GCP Panels for IKONOS “Tyvek” panels Aligned North-South. Control point is center square Lessons Learned Tests with 3 observers agreed within 0.05 m RMS. Narrow gap sometimes hard to see End panels unnecessarily large – Hard to deploy – Too bright See revised design at right References Dennis Helder, Michael Coan, & Peter Gaska, “IKONOS Geometric Characterization” RSE Michael Coan, telecon & . Original Revised
© 2003 Space Imaging, LLC. Confidential and proprietary. Multi-Point Measurements Reduce GCP Pick Error Multiple Point Measurements Average errors of individual points. Symmetric distribution eliminates bias at contrast edge. Accuracy ~ 0.2 pixels Reference C.S.Fraser, H.B.Hanley & T.Yamakawa, “Sub-metre Geopositioning with IKONOS Geo Imagery, ISPRS Joint Workshop “High Resolution Mapping from Space 2001”, Hanover, Germany September
© 2003 Space Imaging, LLC. Confidential and proprietary. Summary & Conclusions Metric Test Range Applications Requirements Verification Calibration of Interior & Exterior Orientation Verification of System Accuracy Monitoring of System Accuracy Test Range Considerations High probability of cloud-free skies Availability of geodetic control Overall size suitable for sensor Cultural features suitable for ground control Ground Control Points Use centerline, not corner features For ultimate accuracy – Deploy panels or use multi-point measurements Thank you.