Georeferencing and Orthoimage Generation from Long Strips...

Institut für Photogrammetrie und GeoInformation

Georeferencing and Orthoimage Generation from Long Strips of ALOS Imagery

Franz Rottensteiner1, Thomas Weser2, Clive Fraser2

1 Institute of Photogrammetry and Geoinformation, Leibniz University Hannover, Germany

2 Cooperative Research Centre for Spatial Information, Department of Geomatics, University of Melbourne, Australia

ALOS Symposium Rhodes, Greece 3. 11. 2008


Outline

• Motivation

• Sensor model and strip adjustment

• Automatic determination of Ground Control Points (GCPs)

• Evaluation of Strip Adjustment

• Conclusion

Outline Motivation Sensor Model and Strip Adjustment GCP Measurement Evaluation Conclusions


Motivation

• Geoscience Australia (GA): Data node for Oceania

• GA want to distribute ortho-rectified ALOS scenes

• Huge demand, e.g. all Western Australia (2.5 million km2)

• Existing solution: processing of individual scenes

• Main bottleneck in the production line: Determination of GCPs

• In order to cope with the huge demand, the output rate has to be increased

Outline Motivation Sensor Model and Strip Adjustment GCP Measurement Evaluation Conclusions

Strip adjustment using a generic sensor model


Generic Pushbroom Sensor Model IOutline Motivation Sensor Model and Strip Adjustment GCP Measurement Evaluation Conclusions

YECSXECS

ZECS

PECS = f(pF,x)

pF

Scene on groundPECS

Satellite path

CCD array records lines at time t = t0 + Δt · yFC

PECS = S(t)PECS = S(t) + RO·RP(t)PECS = S(t) + RO·RP(t) · [CM + RM] PECS = S(t) + RO·RP(t) · [CM + λ· RM·(pF – cF + δx )]PECS = S(t) + RO·RP(t) · [CM + λ· RM·(pF – cF + δx )]

Systematic error correction: δx

PECS = S(t) + RO·RP(t) · [CM + λ· RM·(pF – cF + δx )]


“real” orbit

Approximated orbit

ΔB

ΔB

ΔB

• Orbit observations for path S(t) and attitudes RP(t), affected by systematic errors

• Correction of systematic errors in path and attitudes– S(t) and attitudes RP(t) modelled by cubic splines– Image observations relate to “real” orbit path and attitudes– Observed orbit points are direct observations for S(t) and RP(t) – Unknown bias correction ΔB modelled as time-constant offset

(for path and attitudes)• Bundle adjustment using ground control

points and orbit observations to determine the orbit parametersand the bias corrections

Generic Pushbroom Sensor Model IIOutline Motivation Sensor Model and Strip Adjustment GCP Measurement Evaluation Conclusions


Generic Pushbroom Sensor Model IIIOutline Motivation Sensor Model and Strip Adjustment GCP Measurement Evaluation Conclusions

• Systematic error correction in image space (δx):

• Coefficients can be determined in adjustment ⇒ self-calibration

δx = a0 + a1 · xF + a2 · xF2

δy = b0 + b1 · xF + b2 · xF2

YC

XC

ZC

C

XF4XF3XF2XF1

YF4YF3YF2

YF1

ZF4ZF3ZF2ZF1

• ALOS PRISM:– One set of coefficients per CCD (level 1)– Calibration data from JAXA– Preprocessing: merge sub-images

• ALOS AVNIR-2:– Only one set of coefficients– Preprocessing of level 1 images to get multi-spectral image

YC

XC

ZC

C

XFYF

ZF


Strip AdjustmentOutline Motivation Sensor Model and Strip Adjustment GCP Measurement Evaluation Conclusions

Pushbroom scanner scene 2Orbit PathOrbit AttitudesCamera MountingCamera

Pushbroom scanner scene 1Orbit PathOrbit AttitudesCamera MountingCamera

Camera MountingCamera

Orbit PathOrbit Attitudes

Orbit PathOrbit Attitudes

• Modularised sensor model• Individual scenes can share components

- Internal camera parameters- Exterior orientation ⇒ Strip adjustment

Merged Orbit Attitudes

Merged Orbit Path

• One set of EO parameters

• One set of bias correctionparameters per strip

• Bridging of sceneswithout ground control




Automatic Generation of GCPsOutline Motivation Sensor Model and Strip Adjustment GCP Measurement Evaluation Conclusions

• Work flow:1) Extraction of points from digital orthophoto (Foerstner

interest operator), interpolation of heights in DEM ⇒ 3D GCP candidates

2) Automatic measurement of GCP candidates in images

a) Extraction of image patches from digital orthophoto

b) Maximum cross correlation coefficient (CCC)

c) Least squares matching

d) Eliminate GCP candidates with CCC < threshold

3) Bundle adjustment with robust estimation

• Input data: Existing digital orthophoto + DEM


Strip Adjustment – PRISM: Test setupOutline Motivation Sensor Model and Strip Adjustment GCP Measurement Evaluation Conclusions

• Strip consisting of 10 level 1B1 nadir scenes

• Strip length: 293 km

• 123 GPS points (accuracy < 1m)

• Digital orthophoto from Landsat 7

• DEM from Dept. of Lands NSW

• Four different scenarios for adjustment:

1) Individual orientation (4 GCPs / scene)

2) Strip adjustment (8 GCPs)

3) Strip adjustment (4 GCPs)

4) Strip adjustment (automatic definition of GCPs)

Canberra

5828_1

5789_2

5789_3

5789_4

4922_2

4922_3

4922_4

4922_5

5808_1

5808_2


Strip Adjustment – PRISMOutline Motivation Sensor Model and Strip Adjustment GCP Measurement Evaluation Conclusions

• Altogether 205 GCPs• Computation time: 4:40 minutes

Distribution of automatically determined GCPs


Strip Adjustment – PRISM: ResultsOutline Motivation Sensor Model and Strip Adjustment GCP Measurement Evaluation Conclusions

4922_4

PRISM, RMSX

0.40

0.50

0.60

0.70

0.80

0.90

1.00

1.10

1.20

1.30

1.40

5828_1 5789_2 5789_3 5789_4 4922_2 4922_3 4922_4 4922_5 5808_1 5808_2

Individual OrbitsStrip Adjustment, 8 GCPsStrip Adjustment, 4 GCPsStrip Adjustment, matching

PRISM, RMSY

0.30

0.50

0.70

0.90

1.10

1.30

1.50

1.70

1.90

2.10

2.30

5828_1 5789_2 5789_3 5789_4 4922_2 4922_3 4922_4 4922_5 5808_1 5808_2

Individual OrbitsStrip Adjustment, 8 GCPsStrip Adjustment, 4 GCPsStrip Adjustment, matching

RMS errors at check points• Strip adjustment, 8 GCPs: subpixel accuracy, 65% fewer GCPs• Strip adjustment, 4 GCPs: pixel-level accuracy, 82% fewer GCPs• Automated orientation: Systematic offset of 1/3-1/2 (LS 7) pixels


Strip Adjustment – PRISM: Long stripOutline Motivation Sensor Model and Strip Adjustment GCP Measurement Evaluation Conclusions

• Strip from previous example expanded to 21 nadir scenes

• Strip length: 600 km

• Check points from previous example

• Adjustment with 8 GCPs:

RMS at check points: 0.6 pixels

Maximum RMS: 0.8 pixels

Reduction of GCPs: 81%

• Adjustment with 4 GCPs:

RMS at check points: 0.7 pixels

Maximum RMS: 0.9 pixels


Strip Adjustment – AVNIR: Test setupOutline Motivation Sensor Model and Strip Adjustment GCP Measurement Evaluation Conclusions

• Strip consisting of 4 AVNIR-2 scenes (level 1B1)• Strip length: 250 km• Same points as for PRISM• Digital orthophoto from Landsat 7• DEM from SRTM • Calibration of camera parameters• Three different scenarios for adjustment:

1) Individual orientation (4 GCPs/scene)2) Strip adjustment (4 GCPs)3) Strip adjustment (automatic

definition of GCPs)

Canberra

4924_1

4924_2

4924_3

4924_4


Strip Adjustment – AVNIROutline Motivation Sensor Model and Strip Adjustment GCP Measurement Evaluation Conclusions

• Altogether 148 GCPs• Computation time: 50 seconds

Distribution of automatically determined GCPs


Strip Adjustment – AVNIR: ResultsOutline Motivation Sensor Model and Strip Adjustment GCP Measurement Evaluation Conclusions

RMS errors at check points

• Strip adjustment, 4 GCPs: subpixel accuracy, 65% fewer GCPs• Automated orientation: pixel-level accuracy• Systematic offset relative to LS7 can also be observed


ConclusionOutline Motivation Sensor Model and Strip Adjustment GCP Measurement Evaluation Conclusions

• Strip adjustment can achieve a reduction of the number of GCPs by up to 80% without loss of accuracy

• With PRISM, a 600 km strip could be georeferenced with sub-pixel accuracy using only 8 GCPs

• The accuracy of automated GCP measurement is limited by the quality of the original orthophoto

• Strip adjustment is used for production at Geoscience Australia⇒ Increase in productivity of 300%


ISPRS Hannover Workshop 2009

High-Resolution Earth Imaging for Geospatial InformationHannover, June 2nd - 5th, 2009

hosted by sponsored by

in conjunction with AGILE 2009www.ipi.uni-hannover.de/ipi-workshop.html

Georeferencing and Orthoimage Generation from Long Strips...

Documents

Transcript of Georeferencing and Orthoimage Generation from Long Strips...