18
CMRT09: Object Extraction for 3D City Models, Road Databases and Traffic Monitoring - Concepts, Algorithms, and Evaluation
The best orientation for a road is chosen by minimizing the num
ber of extracted segments (as shown in Fig.9); a road can be de
fined as the minimum set of segments with a length greater than
T2 and same angular value. The set of segments which forms
a road is created applying a clustering algorithm; the DBSCAN
(Ester et al., 1996) is adopted to group the set of extracted seg
ments. A segment belongs to a cluster if and only if the distance
between the initial point of segment and the nearest neighbour is
under a threshold; if this geometric criteria is satisfied the lengths
of clusterized segments are also checked. If the length are compa
rable (in terms of distance from the mean value of the cluster) the
set of cluster is labelled as road and the centerline is calculated.
In Fig. 10 a series of tests on Mannheim data-set for different
orientations is shown. Tests put in evidence that the algorithm,
owing to the clustering, does not consider incoherent segments
(Fig. 10c).
a) b) C)
Figure 10: Road extraction for three different angles; segments
are the thick red lines (bottom), while raw ones are shown in top
The extracted geo-referenced and vectorial road graph with the
proposed technique is shown in Fig.l 1; some roads are not cor
rectly identified due to presence of high density canopies.
Figure 11 : Road graph for Mannheim data-set
5 CONCLUSIONS AND FUTURE WORK
In this paper, we presented a complete methodology to solve the
problem of automatic extraction of urban objects from multi
source aerial data. The procedure, which consists of sequential
steps, takes advantage of classified data with a powerful machine
learning algorithm as AdaBoost with CART as weak learner. The
capability of distinguishing among four classes in an urban area
as Mannheim increases the set of possible applications; two test
cases were presented: building and road extraction. In the case of
building extraction, the fusion of spectral data with LiDAR data
using AdaBoost overtakes the limits of a simple nDSM thresh
olding especially when canopies have a high density. The pro
posed road extraction method allows to reduce the effect of oc-
clusions;roads, extracted with the “line growing” approach en
hanced with clustering, well match with a photo-interpretation
process. As future works, more tests on more complex data with
curved lines will be performed; moreover different weak learners
based on RBF Neural Networks will be tested.
ACKNOWLEDGMENT
We would like to thank C. Nardinocchi and K. Khoshelham at
Delft University of Technology for their helpful support.
REFERENCES
Bacher, U. and Mayer, H., 2005. Automatic road extraction from
multispectral high resolution satellite images. Proceedings of
CMRT05.
Breiman, L., Friedman, J., Olshen, R. and Stone, C., 1984. Clas
sification and Regression Trees. Wadsworth and Brooks, Mon
terey, CA.
Clode, S. P„ Rottensteiner, F. and Kootsookos, P., 2005. Data
acquisition for 3d city models from lidar extracting buildings and
roads. Proceedings of CMRT05.
Duda, R. O., Hart, P. E. and Stork, D. G., 2000. Pattern Classifi
cation. Wiley-Interscience Publication.
Ester, M., peter Kriegel, H., S, J. and Xu, X., 1996. A density-
based algorithm for discovering clusters in large spatial databases
with noise. In: Proceedings of the Knowledge Discovery and
Data Mining Conference, AAAI Press, pp. 226-231.
Frontoni, E„ Khoshelham, K., Nardinocchi, C., Nedkov, S. and
Zingaretti, R, 2008. Comparative analysis of automatic ap
proaches to building detection from multisource aerial data. Pro
ceedings of GEOBIA.
Guo, Y., Bai, Z., Li, Y. and Liu, Y., 2007. Genetic algorithm and
region growing based road detection in sar images. Proceedings
of Int. Conference on Natural Computation pp. 330-334.
Hahn, K., Han, Y. and Hahn, H., 2007. Extraction of partially
occluded elliptical objects by modified randomized hough trans
form. In: Proceedings German conference on Advances in Arti
ficial Intelligence, Springer-Verlag, pp. 323-336.
Kass, M„ Witkin, A. and Terzopoulos, D., 1987. Snakes: Active
contour models. Int. J. of Computer Vision 1 (4), pp. 321 -331.
Liu, H. and Motoda, H., 2008. Computational Methods of Fea
ture Selection. Chapman and Hall/CRC.
Lu, Y., Trinder, J. and Kubik, K., 2006. Automatic building de
tection using the dempster-shafer algorithm. Photogrammetric
Engineering and Remote Sensing 72(4), pp. 395^403.
Marikhu, R., Dailey, M. N., Makhanov, S. and Honda, K., 2006.
A family of quadratic snakes for road extraction. Lecture Notes
in Computer Science ACCV 2007 4843, pp. 85-94.
Mena, J., 2003. State of the art on automatic road extraction for
gis update: a novel classification. Pattern Recognition Letters 24,
pp. 3037-3058.
Schapire, R. and Singer, Y., 1999. Improved boosting algorithms
using confidence-rated predictions. Machine Learning 37(3),
pp. 297-336.
Walter, V., 2004. Object-based classification of remote sensing
data for change detection. ISPRS Journal of Photogrammetry
and Remote Sensing 58, pp. 225-238.
Weidner, U. and Forstner, W., 1995. Towards automatic building
extraction from high resolution digital elevation models. ISPRS
J. of Photogrammetry and Remote Sensing 50(4), pp. 38^49.
