International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004
(a) Detected building boundaries (b) Building detection
on 1 by 1.5 km area examples
Figure 7. Detected building boundaries
3.3 Tree crown
For the entire 8 by 8 km area, tree crowns were extracted.
Some isolated individual tree crowns well match with the
optical images (Figure 8 (b)). However, in aggregated tree
clusters, it is not so clear whether the fitted crown ellipse is
correct or not because both the visual image and the DEM are
too low a resolution. Also, overall accuracy including results for
forest areas are not so good when comparing locally detected
tree crowns in the 1.5 by 1 km sub area (Figure 8 (a)). That’s
because of the local variation in spectral signature.
Po TE P GEHE
(b) Subsets of 8 by 8 km
tree detection results
i. Wf d TH A = gb: À
(a) Tree detection in a 1 by 1.5 km
area
Figure 8. Tree detection in a 1 by 1.5 km area
4. SUBSIDENCE RISK MAPPING
Kelvin (2003) used the tree crown centre locations as detected
here together with the OS® MasterMap® digital map data and
the surface and sub-surface geology from the BGS DiGMapGB-
50 dataset to assess the impact of tree locations on potential
building subsidence. Within the UK, most domestic subsidence
occurs on shrinkable clay soils. Vegetative desiccation is a
prime cause of clay shrinkage. When allied with the
aforementioned commercial datasets, the tree location data
allows mapping of the desiccation zones. If a simple risk
classification is adopted, a relatively straightforward method of
identifying potential occurrences of subsidence can be used
(Figure 9). This offers significant improvements over existing
subsidence risk assessment methodologies, such as those carried
out for insurance purposes, which do not fully account for the
spatial distribution of the causative factors. This method offers
much potential for future refinement. The publication of more
detailed studies of vegetative desiccation characteristics would
830
improve the potential accuracy of the findings, and the
possibility exists of adding value to the OS® MasterMap® data
in the form of information relating to building age and condition.
It is to be hoped that this method will generate results consistent
with those of existing techniques, such as PSInSAR. If this is
the case, then the techniques developed within this paper would
be applicable for predictive purposes within the context of
urban development.
Figure 9. Subsidence Riskmap accounting for buildings
influenced by two or more dessication zones using
tree data derived using the methods described here.
(taken from Figure 4.11 of Kelvin, 2003). Geology
data: IPR/43-39C British Geological Survey. ©
NERC. MasterMap data: Ordnance Survey ©Crown
Copyright.
5. CONCLUSION
A general landscape object detection method for a very large
urban areas has been demonstrated here. It consists of focusing
by DTM construction, refinement of ROI by data fusion
between 3D range data and multi-spectral signature and object
identification by boundary generalization and fitting.
Assessments of final products — DTM , building and tree crown,
shows acceptable quality compared with ground truth like GIS
data sets and appears to be reliable under visual inspection
considering the limited resolution of the 3D range data.
However, object detection quality should be able to be upgraded
with the introduction of more reliable machine vision
algorithms such as a robust generalization process for building
boundaries and splitting methods for tree crown reflectance
using the optical image rather than the Lidar DEM.
Domestic subsidence is a well-known problem when buildings
with shallow foundations (in the UK much of the housing stock
is over 100 years old) lies on clay soils with large amounts of
tree cover. Central to this is the ability to determine individual
tree locations and their proximity to buildings. Using external
information on (a) soil geology; (b) building age; (c) tree root
damage potential depending on tree type; (d) tree height
(determined from lidar), landscape object detection result in this
research can provide a map of buildings under risk from
subsidence.
ACKNOWLEDGEMENTS
The authors would like to thank BNSC and Infoterra Ltd for
supporting this research under the LINK programme as well as
providing the Lidar data-set. We also thank Nicke Coote of the
Ordnance Survey for kindly providing the OS& MasterMap®
data.
REFERENCES
Adams, R., L. Bischof, L., 1994, Seeded Region Growing, /EEE
Transaction on Pattern Analysis and Maschine Intelligence,
16(6), pp. 641 — 647.
Inter
Axel
algo
& Re
Balt:
geon
builc
Resc
21S
Bren
in ur
Rem
Chat
For |
Lette
Fras
geop
ISP}
Hark
Groc
App!
Con}
Groc
Proc
from
Gon,
Mea:
Phot
pp.L
Hers
Peuc
Inter
Kelv
subs
Lonc
Lin,
from
Und
Loor
new
187-
Maa:
extra
Jour.
pp. 1.
Kim.
appr
17.1
Kim.
