1. September 2013, Fontainebleau, France
13th International Scientific and Technical Conference From Imagery to Map: Digital Photogrammetric Technologies
A Case Study of Pleiades Tri-Stereo Imagery:
accuracy assessment, interpretability, 3D modeling potential.
Elena Kobzeva, Chief Engineer, Technology 2000
Nadezhda Malyavina, Head of Racurs Production department
Petr Titarov, Software developer, Racurs
September 2013, Fontainebleau, France

2. A Case Study of Pleiades Tri-Stereo Imagery
Contents
Pleiades imagery orientation accuracy assessment
3D modeling of urban area (the city of Yekaterinburg)
Creating and updating topographic maps using Pleiades imagery

3. Pleiades imagery orientation accuracy assessment
Pushbroom imagery orientation models
Test dataset description
Pleiades imagery orientation accuracy
Rigorous, rational polynomial (RPC) and universal pushbroom models
Pleiades Tri-Stereo product and ground points set
Orientation accuracy of single Pleiades images, stereopairs and the triplet

4. Pushbroom imagery orientation models
Rigorous
Universal
Replacement

5. Rigorous orientation model
Ray reconstruction:
Ray vertex
(sensor position):
Viewing direction:

6. Rational polynomial model (RPC)
where
are polynomials:
The coordinates in the RPC formulae are normalized to fall into the range of [-1;1].

7. Rational polynomial model (RPC) refinements
RPC adjustment: bias removal
RPC adjustment: affine refinement

8. Universal pushbroom models
Parallel-perspective model
Direct Linear Transformation (DLT)
Affine model

9. Test dataset description
Pleiades Tri-Stereo Imagery
Parameters
Images
Image ID
DS_PHR1A_ _ SE1_PX_E060N56_0920_01800
DS_PHR1A_ _ SE1_PX_E060N56_0920_01876
DS_PHR1A_ _ SE1_PX_E060N56_0920_01876
Imaging date and time
:19:53.4
:20:16.6
:20:27.4
Viewing angle along track
10.1°
-2.7°
-8.5°
Viewing angle across track
1.4 °
1.9°
2.1 °

10. Test dataset description
Pleiades Tri-Stereo Imagery – Bundle Product
Pan Image, GSD 0.7 m
Pan Image, GSD 0.7 m
Pan Image, GSD 0.7 m
MS Image, GSD 2.8 m
MS Image, GSD 2.8 m
MS Image, GSD 2.8 m
The images were pan-sharpened using PHOTOMOD

11. Test dataset description
Ground points set
Ground coordinates accuracy: m RMSE
Points measurements in the images accuracy: 1 pixel

12. Pleiades imagery orientation accuracy assessment: methodology
Scheme #
GCPs
number
Orientation model
Objective I
RPC
Supplied RPC accuracy assessment II
RPC + shift
Assessment of accuracy achievable using supplied RPC and tie points (but no ground control)
III 1
Assessment of accuracy achievable with RPC and a single ground control point IV 4
RPC + affine
Assessment of accuracy achievable with RPC and the typical ground control point configuration, applying affine refinement V
To compare the efficiency of affine and shift RPC refinements VI 10
To find out if the accuracy improves with increasing the number of ground control points in the case of applying shift RPC refinement
VII
To find out if the accuracy improves with increasing the number of ground control points in the case of applying affine RPC refinement
VIII
all available
Assessment of the best achievable accuracy in the case of applying shift RPC refinement IX
Assessment of the best achievable accuracy in the case of applying affine RPC refinement X
Affine
Assessment of accuracy achievable with the affine universal model and a minimal set of ground control points, and comparison with orientation with RPC (the ground control points set was the same as in Schemes III and IV). XI
Parallel-perspective
Assessment of accuracy achievable with the various universal models and comparison with orientation with RPC (the ground control points set was the same as in Schemes V and VI).
XII
DLT
XIII

13. Pleiades imagery orientation accuracy: single images
Image phr1a_p_ _sen_
Scheme
GCPs count
Orient. model
GCP RMSE, m
GCP MAX, m
CPs count
CP RMSE, m
CP MAX, m I
RPC - 33
3.1
4.9
III 1
RPC+shift
0.0 32
1.0
2.3 IV 4
RPC+affine
0.4
0.5 29
1.9 V
0.6
0.8
2.0 VI 10
0.7 23
1.1
VII
VIII
2.2 IX
0.9 X
affine
2.6
4.2 XI
par.persp.
1.3
3.9
XII
DLT
1.2
1.7
3.0
XIII
1.6
2.4
1.8
3.4

14. Pleiades imagery orientation accuracy : single images
Image phr1a_p_ _sen_
Scheme
GCPs count
Orient. model
GCP RMSE, m
GCP MAX, m
CPs count
CP RMSE, m
CP MAX, m I
RPC - 38
3.9
4.6
III 1
RPC+shift
0.0 37
0.8
1.5 IV 4
RPC+affine
0.1
0.2 34
0.7
1.7 V
0.3
0.4 VI 10
0.6
0.9 28
1.3
VII
0.5
VIII
1.4 IX
1.2 X
affine
4.5
7.6 XI
par.persp.
2.0
4.1
2.5
4.4
XII
DLT
3.0
XIII
2.8
4.8
5.7

15. Pleiades imagery orientation accuracy : single images
Image phr1a_p_ _sen_
Scheme
GCPs count
Orient. model
GCP RMSE, m
GCP MAX, m
CPs count
CP RMSE, m
CP MAX, m I
RPC - 38
4.2
5.4
III 1
RPC+shift
0.0 37
0.9
1.8 IV 4
RPC+affine
0.1
0.2 34
0.8
2.1 V
0.3 VI 10
0.5 28
0.7
VII
1.9
VIII IX
0.6
1.7 X
affine
6.6
11.3 XI
par.persp.
2.8
5.8
3.7
6.3
XII
DLT
2.6
2.3
4.7
XIII
3.9
6.5
4.0
7.9

16. Pleiades imagery orientation accuracy: single images
Conclusions:
Planimetric accuracy of supplied RPC was RMSE m (the specification is CE90 = 8.5 m).
The accuracy of m RMSE (i.e. rather close to the limit set by the measurements accuracy) was achieved with a single GCP, applying shift refinement to the supplied RPC model.
The accuracy of m RMSE was achieved with 4 GCPs, applying either shift of affine RPC refinement.
Further increasing the number of GCPs did not improve the accuracy.
The orientation accuracy achieved with universal methods varied over a wide range and was significantly worse than one achieved with RPC and bias removal.

17. Pleiades imagery orientation accuracy: the triplet
Triplet orientation without tie points
Scheme
GCPs count
Orient. model
GCP RMSE, m
GCP MAX, m
CPs count
CP RMSE, m
CP MAX, m dS dZ I
RPC - 38
3.8
2.2
4.9
III 1
RPC+shift
0.0 37
0.8
2.3
1.3
5.3 IV 4
RPC+aff.
0.1 34
2.0
4.3 V
0.3
1.6
0.5
2.1
0.7
1.5
4.4 VI 10
0.6
1.4
1.1
2.5 28
VII
RPC+affine
1.0
2.7
2.4
1.8
5.0
VIII
5.9 IX
1.2
6.1 X
affine
13.5
51.4
36.9
125.6 XI
par.persp.
11.5
5.5
23.7
3.4
13.7
7.3
34.3
XII
DLT
0.9
3.1
1.9
19.1
49.3
XIII
10.3
5.1
21.4
3.5
12.9
7.2
30.1

18. Pleiades imagery orientation accuracy: the triplet
Triplet orientation with tie points
Scheme
GCPs count
Orient. model
GCP RMSE, m
GCP MAX, m
CPs count
CP RMSE, m
CP MAX, m dS dZ II
RPC+shift - 38
3.6
2.2
4.5
5.7
III 1
0.2
0.6 37
0.7
1.4
5.6 IV 4
RPC+affine
0.1 34
0.8
2.0
4.3 V
0.3
1.6
0.5
2.1
1.5
4.7 VI 10
1.1
2.4 28
2.3
5.4
VII
5.2
VIII
1.2
5.0 IX X
affine
0.0
8.0
38.0
19.3
67.8 XI
par.persp.
28.7
9.5
64.2
31.9
12.6
95.2
XII
DLT
30.5
7.4
67.3
34.6
11.1
106.2
XIII
30.9
70.5
33.8
13.6
105.8

19. Pleiades imagery orientation accuracy: the triplet
Conclusions:
Using supplied RPC and no GCPs, the achieved planimetric accuracy was 3.6 m RMSE in the case of involving tie points and 3.8 m without them; the vertical accuracy was 2.2 m in both cases. So involving tie points in the adjustment procedure did not significantly improve the accuracy;
Involving GCPs made the difference between adjustment with and without tie points insignificant.
The accuracy of m RMSE was achieved with 4 GCPs, applying either shift of affine RPC refinement. Using a single GCP and applying shift RPC refinement, the planimetric accuracy of m and the vertical accuracy of m were achieved. Increasing GCPs number to 4 allowed improving the results but not significantly, the vertical accuracy became of m.
Further increasing the number of GCPs did not improve the accuracy.
The universal methods are not suitable for stereoscopic (three dimensional) processing of Pleiades imagery.

20. Pleiades imagery orientation accuracy: stereopairs vs. the triplet
Triplet orientation (maximum B:H=0.37)
Scheme
GCPs count
Orient. model
GCP RMSE, m
GCP MAX, m
CPs count
CP RMSE, m
CP MAX, m dS dZ I
RPC - 25
3.6
2.0
4.5
4.9
III 1
RPC+shift
0.0 24
0.8
2.2
1.2
5.3 IV 4
RPC+affine
0.1 21
0.7
1.1
4.3 V
RPC+ shift
0.3
1.6
0.5
2.1
1.9
1.4
4.0 VI 10
0.6
1.3 15
0.9
VII
4.4

21. Pleiades imagery orientation accuracy: stereopairs vs. the triplet
Forward + backward stereopair orientation (B:H=0.37)
Scheme
GCPs count
Orient. model
GCP RMSE, m
GCP MAX, m
CPs count
CP RMSE, m
CP MAX, m dS dZ I
RPC - 25
3.5
1.9
4.5
4.3
III 1
RPC+ shift
0.0 24
0.8
2.2
1.3
4.9 IV 4
RPC+affine
0.1
0.2 21
0.7
2.0
1.1
4.7 V
0.3
1.7
0.6
1.4
3.6 VI 10
1.2 15
1.0
3.9
VII
0.5
2.1
4.0

22. Pleiades imagery orientation accuracy: stereopairs vs. the triplet
Forward + nadir stereopair orientation (B:H=0.25)
Scheme
GCPs count
Orient. model
GCP RMSE, m
GCP MAX, m
CPs count
CP RMSE, m
CP MAX, m dS dZ I
RPC - 25
3.3
2.6
4.2
7.7
III 1
RPC+shift
0.0 24
0.8
2.8
1.4
7.8 IV 4
RPC+ affine
0.6 21
0.7
2.2
1.2
6.4 V
0.4
1.6
0.5
2.5
1.3
6.3 VI 10
1.5
1.0
2.4 15
2.7
6.7
VII
6.5

23. Pleiades imagery orientation accuracy: stereopairs vs. the triplet
Nadir + backward stereopair orientation (B:H=0.11)
Scheme
GCPs count
Orient. model
GCP RMSE, m
GCP MAX, m
CPs count
CP RMSE, m
CP MAX, m dS dZ I
RPC - 25
4.1
3.5
4.9
9.4
III 1
RPC+shift
0.0 24
1.0
3.0
1.7
8.0 IV 4
RPC+ affine
0.2
1.8
0.3
2.5 21
0.7
3.4
1.5
10.1 V
0.4
2.7
0.6
4.0
3.3
1.2
8.8 VI 10
2.4
4.6 15
1.4
VII
0.5
2.2
4.3
2.9
7.8

24. Pleiades imagery orientation accuracy: stereopairs vs. the triplet
Conclusions:
The accuracy of orientation of the triplet and of the forward+backward stereopair (i.e. the stereopair with the largest base-to-height ratio) was approximately the same.
The accuracy of triplet orientation was slightly better than one of the stereopairs which included the nadir image (so the stereopairs had lower base-to-height ratio).

25. Mapping and 3D modeling of urban areas
Creating 3D models
Deriving DEM
Generating orthoimagery
3D modeling of urban area
Assessment of suitability for topographic maps creating and updating
Interpretability assessment
Assessment of objects positioning accuracy
Drawing contour lines

26. Creating 3D models using PHOTOMOD: deriving DEM

27. Creating 3D models using PHOTOMOD: generating orthoimagery

28. Creating 3D models using PHOTOMOD: 3D vectorization

29. Creating 3D models using PHOTOMOD: automatic 3D modeling

30. PHOTOMOD. Object texturing

31. PHOTOMOD. Model texturing using close-range imagery

32. PHOTOMOD. Import of “special” objects

33. PHOTOMOD. Creating 3D model of the city of Yekaterinburg

34. Interpretability assessment
Source dataset:
Pleiades orthoimagery, 0.5m, RGB
Worldview-2 orthoimagery, 0.5m, RGB
A3 orthoimagery, 0.1m, RGB
Topographic interpretation samples set WV-2, GE-1 and Ikonos
Scanned topographic plans of scale 1:500, contour interval 0,5 m;
Vector topographic maps of scale 1: , contour interval 2 m

35. Interpretation results
Imagery
Number of recognized objects
Images only
Add. Info
Field ve-rification
Not re-cognized
Pleiades
103 92 21 20
WV 2
106 90 19 A3
126 78 15 14

36. Assessment of objects positioning accuracy
Source dataset:
Pleiades stereopair (9º and -11º), 0,5 m, PAN
Pleiades orthoimagery 0.5m, RGB
WV2 orthoimagery, 0.5m, RGB
A3 orthoimagery 0.1m, RGB as reference data

37. PHOTOMOD. Comparing different types of objects
Single-storey private houses
Multistory city buildings
Pleiades A3
Pleiades A3

38. Interpretability analysis
It is impossible to tell residential buildings from nonresidential ones
1.5 m - wide ledges are indiscernible
Shape and size of multistory buildings are reconstructed correctly
Some architectural forms may be missing (the ledges are
shown on one side of the building and missing on the other)

39. Assessment of objects positioning accuracy
Parameter
WV2 ortho
Pleiades ortho
Pleiades stereo
Number of measurements
371
366
Mean error, m
1.3
1.5
0.9
Maximum error, m
4.0
5.4
3.5
Error distribution – vector map of scale 1: 2 000
Error distribution – vector map of scale 1 : 5 000
0-0,4 mm mm
larger than 0.8 mm
0-0,4 mm mm
larger than 0.8 mm

40. PHOTOMOD. Drawing contour lines

41. Contour lines verification using reference data
Vector topographic maps of scale 1:10 000,
contour interval 2 m
Contour lines derived from the Pleiades stereopair

42. Topographic mapping and 3D modeling of urban areas
Conclusions:
The 3D model created is geometrically accurate and discrete, so it is possible to access separate objects, to set attribute values for them and to perform 3D measurements - in other words, to produce geospatial databases. The model can be used for visualization and for 3D city planning. Stereoscopic measurements ensure better accuracy and interpretability than ones performed in single images, while using tri-stereo imagery reduces “blind zones”.
Pleiades images are suitable for creating and updating topographic maps of scale up to 1: If additional sources of data are available and field verification is possible, it is possible to create and update 1 : scale maps of moderate-sized inter-settlement areas.
Accuracy and interpretability of Pleiades imagery are comparable to ones of WordView-2.

43. RACURS and TECHNOLOGY 2000 ASTRIUM GeoInformation Services
Acknowledgement
RACURS and TECHNOLOGY 2000
express their gratitude to
ASTRIUM GeoInformation Services
for the Pleiades Tri-Stereo Imagery Product
over the city of Yekaterinburg

44. A Case Study of Pleiades Tri-Stereo Imagery
Thank you for attention !