OrangeNXT - High accuracy mapping from videos for efficient fiber optic cable...
10008-16.antoine_lefebvre2
1. E X T R A C T I O N O F
U R B A N V E G E T A T I O N
W I T H P L É I A D E S
M U L T I - A N G U L A R I M A G E S
Antoine Lefebvre*, Jean Nabucet, Thomas Corpetti, Nicolas Courty,
Laurence Hubert-Moy
* CNES, UMR IRISA
2. C O N T E X T
U R B A N V E G E T A T I O N
Monitoring urban vegetation,
WHY ?
• Climate adaptation (UHI)
• Health & Quality of life
Copernicus Urban Atlas!
Copernicus Urban Atlas Street Tree!
Existing data
• European level: Street Tree Layer
• National level: IGN BD TOPO
• Minimal Mapping Unit: 500m2
3. C O N T E X T
M U L T I - A N G U L A R I M A G E S
Pléiades’ stereo and tistereo
capability ables to generate
High Resolution
Digital Elevation Model
How can it help to extract
urban vegetation ?
4. S T U D Y A R E A & D A T A
Rennes City, France
• 2,500 km2
• 200,000 inhabitants
• CNES Kalideos Program
Pair of Pléiades images:
5. M E T H O D
1.
nDSM
generation
2.
Spectral &
Contextual
features
computation
3.
Bagging Tree
classification
6. M E T H O D
N D S M G E N E R A T I O N
Digital Soil Model (DSM)
generated with ERDAS eATE
• User friendly
• Efficient on large areas
Normalized-DSM (nDSM)
Extraction of surface features
with SAGA GIS (slope based
method)
7. M E T H O D
F E A T U R E S C O M P U T A T I O N
Spectral information :
- NDVI
- NDWI
Contextual information (1/2) :
Texture
- GLCM and Haralick’s indexes
à 10 directions
à Contrast, Energy, Entropy
- Rotation invariant features
à min, max
8. M E T H O D
F E A T U R E S C O M P U T A T I O N
Contextual information (2/2) :
Granulometry
- Differential Mophological
Profiles (DMP)
9. M E T H O D
B A G G I N G T R E E C L A S S I F I C A T I O N
Decision tree avantage:
- Easily ignore redundant variables
Bagging tree concept & avantages:
- Fit many large trees and classify by majority vote
- Reduce the variance of unstable trees, leading to
improved prediction
10. 1. Multispectral (MS)
2. Multispectral + nDSM (MS + DSM)
3. Multispectral + Contextual (MS + DSM + CON)
4. Multispectral + Contextual + nDSM (MS + DSM + CON)
M E T H O D
F E A T U R E S C O M B I N A T I O N
14. R E S U L T S
Rate of omission and commission errors
nDSM
contributes to
decrease
commission errors
15. R E S U L T S
nDSM does not
help to extract
small objects
Rate of omission and commission errors
Omission errors for 5 high vegetation classes
nDSM
contributes to
decrease
commission errors
17. R E S U L T S
Bagging tree variables importance
18. R E S U L T S
C O M P A R I S O N W I T H E X I S T I N G
D A T A
IGN BD TOPO Classification IGN BD TOPO Classification
Public garden University campus
19. R E S U L T S
C O M P A R I S O N W I T H E X I S T I N G
D A T A
IGN BD TOPO Classification IGN BD TOPO Classification
Suburban housing Downtown housing
20. C O N C L U S I O N & F U T U R E
W O R K S
• Evaluation of Pléiades multi-
angular images to extract urban
vegetation
• Complementary with spectral and
contextual information
• nDSM reduces commission errors
• First results have higher precision
than existing data
• Application to tri-stereo Pléiades
images
• 3D model generation with
Orfeo Toolbox
• Characterize high vegetation in
different classes
• Comparison with LIDAR data
21. E X T R A C T I O N O F
U R B A N V E G E T A T I O N
W I T H P L É I A D E S
M U L T I - A N G U L A R I M A G E S
Antoine Lefebvre*, Jean Nabucet, Thomas Corpetti, Nicolas Courty,
Laurence Hubert-Moy
* CNES, UMR IRISA