Filtering of matching data • Filtering process is performed by using

ISPRS Technical Commission I Symposium “From Sensors To Imagery”, Paris – Marne la Vallée, France, 3-6 July 2006. Institute of. Photogrammetry and.
2MB taille 0 téléchargements 57 vues
Institute of Photogrammetry and GeoInformation

Filtering of matching data • Filtering process is performed by using variance and Foerstner roundness criteria in the matching kernel windows

ISPRS Technical Commission I Symposium “From Sensors To Imagery”, Paris – Marne la Vallée, France, 3-6 July 2006

Institute of Photogrammetry and GeoInformation

Filtering of matching data • Filtering process is performed by using variance and Foerstner roundness criteria in the matching kernel windows

ISPRS Technical Commission I Symposium “From Sensors To Imagery”, Paris – Marne la Vallée, France, 3-6 July 2006

Institute of Photogrammetry and GeoInformation

Comparison: Laser DEM and HRS DEM Test area Gars (5 x 5 km): part of Bavarian test site

ISPRS Technical Commission I Symposium “From Sensors To Imagery”, Paris – Marne la Vallée, France, 3-6 July 2006

Institute of Photogrammetry and GeoInformation

Absolute differences and standard deviations Comparison of the regularized SPOT-DEM and the reference DEM for selected areas and two surface types (SPOT DEM – Reference DEM) Reference area

Z-bias [m] non-forest / STDV

Z-bias [m] forest / STDV

Height-difference forest/non-forest

DEM-02

7.8 / 5.6

17.8 / 9.7

11.9

DEM-02 matching filter

8.5 / 4.8

15.0 / 8.0

7.5

DEM-02 median filter (DEM)

8.0 / 3.9

17.2 / 8.2

9.2

DEM-06

6.5 / 6.5

19.0 / 9.0

12.5

DEM-06 median filtering

6.2 / 4.2

16.9 / 6.8

10.7

ISPRS Technical Commission I Symposium “From Sensors To Imagery”, Paris – Marne la Vallée, France, 3-6 July 2006

Institute of Photogrammetry and GeoInformation

Comparison: Laser DEM and HRS DEM Laser

HRS-DEM of Gars Area

HRS

Area: 5 km x 5 km, height levels: 406 to 598 meters ISPRS Technical Commission I Symposium “From Sensors To Imagery”, Paris – Marne la Vallée, France, 3-6 July 2006

Institute of Photogrammetry and GeoInformation

Difference between Laser DEM and HRS DEM

Mean height difference: 6.3 meter (not adjusted orientation), forest: 12 meter, open areas 4.1 meter sigma: 4.8 meter (max. deviation 30 meter) Area: 5 km x 5 km, height levels: 406 to 598 meters ISPRS Technical Commission I Symposium “From Sensors To Imagery”, Paris – Marne la Vallée, France, 3-6 July 2006

Institute of Photogrammetry and GeoInformation

DEM and orthoimage of Catalonia

ISPRS Technical Commission I Symposium “From Sensors To Imagery”, Paris – Marne la Vallée, France, 3-6 July 2006

Institute of Photogrammetry and GeoInformation

Barcelona Test Site (DEM accuracy without GCP) Reference area

Size of Area

Mean Height Difference [m]

STDV [m]

Min / Max

SPOT-HRS-DEM Barcelona City

71 km²

11.2

4.4

-47 / +37

Rural Area

161 km²

10.4

5.3

-59 / +53

Moderate Mountain

105 km²

11.1

6.5

-62 / +63

Montserrat (steep mountains)

84 km²

9.8

13.5

-158 / +191

Whole area

1882 km²

10.0

6.3

-158 / +191

ISPRS Technical Commission I Symposium “From Sensors To Imagery”, Paris – Marne la Vallée, France, 3-6 July 2006

Institute of Photogrammetry and GeoInformation

Comparison of SPOT and SRTM DEM Reference area

Size of Area

bias [m]

STDV [m]

Min / Max

SPOT-HRS-DEM Barcelona City

71 km²

11.2

4.4

-47 / +37

Rural Area

161 km²

10.4

5.3

-59 / +53

Moderate Mountain

105 km²

11.1

6.5

-62 / +63

Montserrat

84 km²

9.8

13.5

-158 / +191

Whole area

1882 km²

10.0

6.3

-158 / +191

Catalonia Area

SRTM-DEM Barcelona City

71 km²

1.0

4.7

-22 / +25

Rural Area

161 km²

-1.5

7.2

-98 / +62

Moderate Mountain

105 km²

-1.4

8.7

-218 / +135

Montserrat

84 km²

-1.8

25.2

-484 / +394

Whole area

623 km²

-1.2

8.5

-484 / +394

ISPRS Technical Commission I Symposium “From Sensors To Imagery”, Paris – Marne la Vallée, France, 3-6 July 2006

Institute of Photogrammetry and GeoInformation

Fusion of SPOT and SRTM DEM Catalonia Area SRTM DEM as reference for mean height Reference area

Size of Area

bias [m]

STDV [m]

Min / Max

Barcelona City

71 km²

0.9

3.7

-23 / +28

Rural Area

161 km²

-1.3

4.9

-62 / +48

Moderate Mountain

105 km²

-1.0

5.6

-70 / +78

Montserrat

84 km²

-1.5

11.1

-183 / +201

Whole area

623 km²

-1.0

5.7

-183 / +201

ISPRS Technical Commission I Symposium “From Sensors To Imagery”, Paris – Marne la Vallée, France, 3-6 July 2006

Institute of Photogrammetry and GeoInformation

Improvement of DEM (SPOT + SRTM) with fusion

forest open area build up

forest open area build up ISPRS Technical Commission I Symposium “From Sensors To Imagery”, Paris – Marne la Vallée, France, 3-6 July 2006

Institute of Photogrammetry and GeoInformation

Fusion of SPOT and SRTM DEM

forest open area build up

forest open area build up ISPRS Technical Commission I Symposium “From Sensors To Imagery”, Paris – Marne la Vallée, France, 3-6 July 2006

Institute of Photogrammetry and GeoInformation

Improvement of exterior orientation Pitch angle and stereo angle corrections estimated from 6 GCP (pitch angle: HRS1: 0.00007º, stereo angle: HRS2: 0.0004º) Reference area

Size and Accuracy of Ref-DEM

Mean Height Difference [m]

Std. Dev. [m] (DEM points)

DEM-01, Prien

5 x 5 km, 0.5 m

1.2

5.1

DEM-02, Gars

5 x 5 km, 0.5 m

1.1

6.3

DEM-03, Peterskirchen

5 x 5 km, 0.5 m

-0.7

5.5

DEM-04, Taching

5 x 5 km, 0.5 m

-0.7

5.4

DEM-05, Inzell

10 x 10 km, 5 m

0.1

7.5

50 x 30 km, 2-3 m

0.4

4.9

DEM-06, Vilsbiburg

ISPRS Technical Commission I Symposium “From Sensors To Imagery”, Paris – Marne la Vallée, France, 3-6 July 2006

Institute of Photogrammetry and GeoInformation

Test area Gars (rolling area) (results IPI Hannover) RMSE [m]

bias [m]

RMSE bias

RMSE as Function of terrain inclination

DSM: all points

10.2

-5.5

8.5

8.7 + 10.6 ∗ tan α

DSM: open areas

6.7

-3.0

5.9

6.4 + 4.9 ∗ tan α

DSM: forest

17.0

-14.3

9.2

16.4 + 3.4 ∗ tan α

DEM: all points

5.7

-2.0

5.1

5.0 + 5.4 ∗ tan α

DEM: open areas

4.4

-1.3

4.1

4.2 + 1.6 ∗ tan α

DEM: forest

12.3

-8.5

8.6

10.0 + 6.9 ∗ tan α

DEM = DSM filtered with Hannover program RASCOR ISPRS Technical Commission I Symposium “From Sensors To Imagery”, Paris – Marne la Vallée, France, 3-6 July 2006

Institute of Photogrammetry and GeoInformation

Test area Inzell (mountainous) (results IPI Hannover) 68% forest, DEM only partly by laser scanning, mainly from map with SZ=+/-5m RMS [m]

bias [m]

RMS bias

RMS as function of terrain inclination

DSM: all points

17.2

-10.3

13.8

10.7 +36.1 ∗ tan α

DSM: open area

9.8

-4.7

8.7

6.1 + 39.6 ∗ tan α

DSM: forest

19.9

-13.0

15.0

17.9 + 8.5 ∗ tan α

DEM: all points

13.7

-7.1

11.7

7.9 + 45.7 ∗ tan α

DEM: open area

6.7

-2.5

6.2

4.8 + 26.1 ∗ tan α

DEM: forest

17.4

-10.8

13.7

16.2 + 7.4 ∗ tan α

DEM = DSM filtered with Hannover program RASCOR white = open area black = forest ISPRS Technical Commission I Symposium “From Sensors To Imagery”, Paris – Marne la Vallée, France, 3-6 July 2006

Institute of Photogrammetry and GeoInformation

Conclusion of DEM generation with SPOT 5 HRS Results HRS-Data of very good quality (image and ancillary data) - two orthoimages match with pixel accuracy, applying a small shift value! Comparison to reference DEM show low standard deviations for single points but still some blunders and artifacts in regularized DEM Absolute accuracy of DEM without GCP is in the order of 10 meter, with few GCP just one/two parameters of exterior orientation can be estimated for absolute height accuracy of 1-2 meter for the DEM Further work Improvement of dense matching (region growing) algorithm by introducing more constraints and object information Improvement of exterior orientation with few ground control points in more cases (which parameters can be / have to be improved?) DEM fusion with SRTM DEMs and other sources ISPRS Technical Commission I Symposium “From Sensors To Imagery”, Paris – Marne la Vallée, France, 3-6 July 2006

Institute of Photogrammetry and GeoInformation

OrbView-3 stereo model in city area, same orbit, h/b = 1.4 quality image of matching OrbView-3 stereo model grey value 255 = correlation coefficient = 1.0 grey value 127 = correlation coefficient = 0.5 grey value 0 = correlation coefficient = < 0.5

problems in city with narrow streets and steep areas, high correlation values in open areas

ISPRS Technical Commission I Symposium “From Sensors To Imagery”, Paris – Marne la Vallée, France, 3-6 July 2006

Institute of Photogrammetry and GeoInformation

IKONOS stereo model, same orbit, h/b = 7.5

Optimal conditions for automatic image matching Nearly all points matched Accuracy of building heights = 1.7m not accepted frequency distribution correlation coefficient

Corresponding to Spx = 0.22 GSD Confirmed by ray intersection Spy = 0.25 GSD

ISPRS Technical Commission I Symposium “From Sensors To Imagery”, Paris – Marne la Vallée, France, 3-6 July 2006

Institute of Photogrammetry and GeoInformation

Conclusion Automatic image matching of very high resolution space images in city areas more difficult like for lower resolution – more strong influence of vertical or inclined objects Images of stereo scene from same day optimal, but 10 days difference in time usually not a problem Vertical accuracy not linear depending upon height to base relation because matching not so accurate for large convergence angle (large base) Optimal height to base relation ~ 3.0, but depending upon object and area Different matching results not caused by different sensors, only caused by different geometric and imaging conditions Only smaller GSD of QuickBird leads to more detailed and more accurate result

ISPRS Technical Commission I Symposium “From Sensors To Imagery”, Paris – Marne la Vallée, France, 3-6 July 2006