CMRT09: Object Extraction for 3D City Models, Road Databases and Traffic Monitoring - Concepts, Algorithms, and Evaluation
(a) Matikainen et al., 2007
(b) Champion, 2007
(e) Rottensteiner, 2008
(f) Champion, 2007
(g) Rottensteiner, 2008
(h) Olsen and Knudsen, 2005
Figure 3. Evaluation Details (same colour code as Figure 1). FP new cases related to DSM errors (shadow areas), in Marseille
streets (a)-(b) and Toulouse (c)-(d); (e)-(f) FN new cases (small changes); (g)-(h) FP new buildings related to bridges.
completeness rates for demolished buildings and the high
correctness for unchanged buildings that could be achieved in
these contexts highlight the effectiveness of the presented
approaches in verifying the existing objects in the databases.
The main limitation in terms of qualitative efficiency concerns
the relatively high number of FN new buildings - up to 12.1%
in the Marseille test area with (Rottensteiner, 2008) - that are
mostly related to the object change size. The economical
efficiency of the presented approaches seems to be promising,
with 80-90% of the existing buildings requiring no further
attention by the operator. These buildings are reported to be
unchanged, which saves a considerable amount of manual
work. In terms of the economical efficiency, the main limitation
is a high number of FP demolished buildings that have to be
inspected unnecessarily. Again, this is mainly caused by
problems in detecting small changes.
Areas of improvement should concern input data and
methodologies. Thus, the resolution of LIDAR data
(1 point/m 2 ) used in this test appeared to be critical for the
change detection performance: using higher density LIDAR
data (e.g. 5-10 points / m 2 ) should improve the situation. As far
as methodology is concerned, new primitives should be used in
the algorithms, in particular 3D primitives (representing e.g. the
3D roof planes or building outlines) that can now be reliably
reconstructed with the 3D acquisition capabilities, offered by
recent airbome/spacebome sensors. Another concern should be
the improvement of the scene models used in object detection
such that they can deal with different object classes and their
mutual interactions. By incorporating different object classes
and considering context in the extraction process, several object
classes could be detected simultaneously, and the extraction
accuracy of all interacting objects could be improved.
In this project, we learned how difficult it is to compare
approaches of very different designs. To carry out a fair test, we
chose to use the building label images and to limit the type of
changes to demolished and new buildings. In addition, we chose
to compare the building label images to the initial vector
database, basing on a coverage rate featured by the parameter
T h . Further investigations are necessary to study the actual
impact of this parameter on the completeness and correctness
rates. However, if we are aware of these drawbacks, we think
that this scheme was sufficient to bring out some interesting
findings. We also hope that our results - in conjunction with
those of e.g. the ARMURS 3 project - will be helpful to create a
nucleus of interested people, both in academia and private
sector, and to speed up the progress in the vector change
detection field.
REFERENCES
Breiman, L., Friedman, J. H., Olshen, R. A., Stone, C. J., 1984.
Classification and regression trees. The Wadsworth Statistics /
Probability Series, Wadsworth, Inc., Belmont, California.
Champion, N., 2007. 2D building change detection from high
resolution aerial images and correlation Digital Surface Models. In:
IAPRSIS XXXVI-3/W49A, pp. 197-202.
Champion, N., Boldo, D., 2006. A robust algorithm for estimating
Digital Terrain Models from Digital Surface Models in dense urban
areas. In: IAPRSIS XXXVI-3, pp. 111-116.
N. Champion, L. Matikainen, F. Rottensteiner, X. Liang, J. Hyyppa,
2008. A test of 2D building change detection methods: Comparison,
evaluation and perspectives. In: IAPRSIS XXXVII - B4, pp. 297-304.
Heipke, C., Mayer, H., Wiedemann, C., Jamet, O., 1997. Automated
reconstruction of topographic objects from aerial images using
vectorized map information. In: IAPRS, XXX11, pp. 47-56.
Kumar, S. and Hebert, M.„ 2006. Discriminative random fields.
International Journal of Computer Vision 68(2), pp. 179-201.
Matikainen, L., Kaartinen, K., Hyyppa, J., 2007. Classification tree
based building detection from laser scanner and aerial image data. In:
IAPRSIS XXXVI, pp. 280-287.
Mayer, H., 2008. Object extraction in photogrammetric computer
vision. ISPRS Journal of Photogrammetry and Remote Sensing
63(2008), pp. 213-222.
Olsen, B., Knudsen, T., 2005. Automated change detection for
validation and update of geodata. In: Proceedings of 6th Geomatic
Week, Barcelona, Spain.
Pierrot-Deseilligny, M., Paparoditis, N., 2006. An optimization-based
surface reconstruction from Spot5- HRS stereo imagery. In: IAPRSIS
XXXVI-1/W41, pp. 73-77.
Rottensteiner, F., 2008 Automated updating of building data bases from
digital surface models and multi-spectral images. In: IAPRSIS XXXVII
- B3A, pp. 265-270.
3 http://www.armurs.ulb.ac.be. Last visited: 30 June 2009.
150
