(geometric accuracy and level of details). Based on the desired
LoD
, exclusivity, and semantic level, an error collection is
considered. Model quality is then predicted using a super-
vised Random Forest classifier. Each model provides intrinsic
geometrical characteristics that are compiled in a handcrafted
feature vector. Remote sensing modalities can be introduced.
This helps better describing building models and detecting
errors. Attributes can indeed be extracted by comparing the
3D model with optical images or depth data at the spatial
resolution at least similar to the input 3D model. Experiments
shows it helps detecting hard cases both for geometrical and
topological errors.
This new framework was applied to the case of aerial urban
reconstruction, where features are extracted from
VHR
airborne
images and a
DSM
. A fully annotated dataset containing 3235
aerial reconstructed building models with high diversity and
from three distinct areas was used to test our method. It was
associated with multimodal Red Green Blue optical and Digital
Surface Model features. Although being mitigated over under-
represented errors, results are satisfactory in the well balanced
cases. More importantly, we proved that the urban scene com-
position affects greatly error detection. In fact, some predic-
tions scores are not only stable, when training on a different
urban scene, they even outperform when learning on the same
scene. Additionally, we reported how, for a heterogeneous
training dataset, the size of the training set have, practically no
effect, as test score stay stable for all errors. This demonstrates
that the proposed framework can be easily scaled with the
right choice of training samples with little manually generated
data. This exactly answers to the raised problematic, contrarily
to the present state-of-the-art literature. As a next step, more
structure-aware features (based on graph comparison, for
instance) could be proposed (Boguslawski
et al.
2011) so as to
be applied on a richer and more diverse dataset (potentially
involving data augmentation) under a deep-based framework.
References
Akca, D., M. Freeman, I. Sargent and A. Gruen. 2010. Quality
assessment of 3D building data.
The Photogrammetric Record
25
(132):339–355.
Berger, M., Levine, Nonato, G. Taubin and Silva. 2013. A benchmark for
surface reconstruction.
ACM Transaction
Biljecki, F., H. Ledoux, X. Du, J. Stoter, Soon
The most common geometric and seman
datasets. Pages 13–22 in
ISPRS Annals of the Photogrammetry,
Remote Sensing and Spatial Information Sciences,
Vol. IV-2/W1,
held in Athens, Greece.
Biljecki, F., H. Ledoux and J. Stoter. 2016b. An improved LOD
specification for 3D building models.
Computers, Environment
and Urban Systems
59: 25–37.
Biljecki, F., H. Ledoux and J. Stoter. 2017. Generating 3D city models
without elevation data.
Computers, Environment and Urban
Systems
64:1–18.
Biljecki, F., J. Stoter, H. Ledoux, S. Zlatanova and A. Cöltekin. 2015.
Applications of 3D city models: State of the art review.
ISPRS
International Journal of Geo-Information
4 (4):2842–2889.
Boguslawski, P., Gold and H. Ledoux. 2011. Modelling and analysing
3d buildings with a primal/dual data structure.
ISPRS Journal of
Photogrammetry and Remote Sensing
66 (2):188–197.
Boudet, L., N. Paparoditis, F. Jung, G. Martinoty and M. Pierrot-
Deseilligny. 2006. A supervised classification approach towards
quality self-diagnosis of 3D building models using digital aerial
imagery.
International Archives of the Photogrammetry, Remote
Sensing and Spatial Information Sciences
36 (3):136–141.
Brédif, M., D. Boldo, M. Pierrot-Deseilligny and H. Maïtre. 2007. 3D
building reconstruction with parametric roof superstructures.
Pages 537–540 in
IEEE International Conference on Image
Processing
, held in San Antonio, Tex., 16–19 September 2007.
Breiman, L. 2001. Random forests.
Machine Learning
45 (1):5–32.
Cabezas, R., J. Straub and Fisher. 2015. Semantically-aware aerial
reconstruction from multi-modal data. Pages 2156–2164 in
IEEE
International Conference on Computer Vision
, held in Santiago,
Chile, 11–18 December 2015.
Criminisi, A. and J. Shotton. 2013.
Decision Forests for Computer
Vision and Medical Image Analysis
. Springer Science & Business
Media.
Dick, A. R., Torr and R. Cipolla. 2004. Modelling and interpretation
of architecture from several images.
International Journal of
Computer Vision
60 (2):111–134.
Duan, L. and F. Lafarge. 2016. Towards large-scale city reconstruction
from satellites. Pages 89–104 in
European Conference on
Computer Vision
, held in Amsterdam, The Netherlands, 8–16
October 2016. Springer.
Durupt, M. and F. Taillandier. 2006. Automatic building
reconstruction from a Digital Elevation Model and cadastral
data: An operational approach.
International Archives of the
Photogrammetry, Remote Sensing and Spatial Information
Sciences
36 (3):142–147.
Elberink, S. O. and G. Vosselman. 2011. Quality analysis on 3D
building models reconstructed from airborne laser scanning
data.
ISPRS Journal of Photogrammetry and Remote Sensing
66
(2):157–165.
Ennafii, O., A. Le Bris, F. Lafarge and C. Mallet. 2019. The necessary
yet complex evaluation of 3D city models: A semantic approach.
In
IEEE/ISPRS Joint Urban Remote Sensing Event
(JURSE), held
in Vannes, France, 22–24 May 2019.
Haala, N. and M. Kada. 2010. An update on automatic 3D building
reconstruction.
ISPRS Journal of Photogrammetry and Remote
Sensing
65 (6):570–580.
Hu, Y., Q. Zhou, X. Gao, A. Jacobson, D. Zorin and D. Panozzo. July
2018. Tetrahedral meshing in the wild.
ACM Transactions on
Graphics
37 (4):1–14.
Jaynes, C., E. Riseman and A. Hanson, 2003. Recognition and
reconstruction of buildings from multiple aerial images.
Computer Vision and Image Understanding
90 (1):68–98.
Kaartinen, H., J. Hyyppä, E. Gülch, G. Vosselman, L. Matikainen,
A. Hofmann, U. Mäder, A. Persson, U. Söderman
et al.
2005.
Accuracy of 3D city models: EuroSDR comparison.
International
Archives of the Photogrammetry, Remote Sensing and Spatial
Information Sciences
36 (3/W19):227–232.
Karantzalos, K. and N. Paragios. 2010. Large-scale building reconstruc-
tion through information fusion and 3-d priors.
IEEE Transactions
Remote Sensing
48 (5):2283–2296.
onka and Mitra. 2017. Bigsur: Large-scale
construction.
ACM Transactions on Graphics
.
Kolbe, T. H., G. Gröger and L. Plümer. 2005. CityGML: Interoperable
access to 3D city models. Pages 883–899 in
Geo-Information for
Disaster Management
. Springer.
Kovashka, A., O. Russakovsky, L. Fei-Fei, K. Grauman
et al.
2016.
Crowdsourcing in computer vision.
Foundations and Trends in
Computer Graphics and Vision
10 (3):177–243.
Kowdle, A., Chang, A. Gallagher and T. Chen. 2011. Active learning
for piecewise planar 3D reconstruction. Pages 929–936 in
IEEE
Conference on Computer Vision and Pattern Recognition
(CVPR),
held in Colorado Springs, Colo., 20–25 June 2011.
Lafarge, F., X. Descombes, J. Zerubia, and M. Pierrot-Deseilligny.
2010. Structural approach for building reconstruction from a
single DSM.
IEEE Transactions on Pattern Analysis and Machine
Intelligence
32 (1):135–147.
Lafarge, F. and C. Mallet. 2012. Creating large-scale city models from
3D-point clouds: A robust approach with hybrid representation.
International Journal of Computer Vision
99 (1):69–85.
Ledoux, H. 2018. Val3dity: Validation of 3D GIS primitives according
to the international standards.
Open Geospatial Data, Software
and Standards
3 (1):1.
Ledoux, H. and M. Meijers. 2011. Topologically consistent 3D
city models obtained by extrusion.
International Journal of
Geographical Information Science
25 (4):557–574.
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
December 2019
877
