is fairly similar to the decimetre accuracy values quoted by
authors such as Lohr (1998).
One of the main error sources for planimetric accuracy was the
planimetric shift observed between the LIDAR data and the 2D
vector building outlines. In some cases the occurrence of roof
edge pixels did not coincide with the building boundary and
was often up to one pixel width outside that of the boundary.
As Huising and Gomes Pereira (1998) admit this shift may be
due to local factors such as planimetric building errors. The
vector data used is claimed to have a planimetric accuracy of
anything up to +/-1m. Systematic error may also cause a shift
in building information. Shadowing resulting from low LIDAR
scan angles (+/-19 degrees for this study) can be seen
predominantly in the industrial area due to the high building
sides hiding other building sections from the LIDAR sensor. A
comprehensive guide of other error sources from laser scanner
systems can be found in Huising and Gomes Pereira (1998).
4.4 LIDAR and Roof Detail Extraction - Next Steps
The use of relatively low resolution LIDAR data for roof detail
extraction is restricted by the generalisation of roof structure
information into grid cells. Mixed pixel effects created by the
interpolation of the raw point data into a regular grid have
contributed to increasing the difficulty with which roof detail
can be extracted. One solution may be to use the raw data itself
as demonstrated successfully by Maas and Vosselmann (1999).
The processing of point data, however, tends to be a harder
task compared to the simpler algorithms that are needed to
manipulate grid data. Another solution could be to use a data
set with a higher grid resolution derived from a higher point
spacing. Increases in the level of roof detail may be possible
with this higher resolution data, but the increases may not be
sufficient enough to justify the higher cost of the data.
Having discussed alternatives to the LIDAR data used in this
study is not to say that it did not produce interesting results.
The results derived from using the aspect parameter from the
LIDAR data showed promising signs of being used as a useful
indicator of the dominant ridges and segments of a building
roof. There is the possibility that aspect could be used to define
sub-ridges by splitting the roof aspect values into more than
two groups. Some initial testing using four groups showed
however that this produced poor results with the aspect groups
imposing their own structure on the roof.
Instead of trying to increase the amount of detail that can be
extracted using this resolution data, a task which is unlikely to
be successful, attention should be concentrated on using the
available results for applications that require general roof
detail. Applications such as virtual reality and the visualisation
aspects of planning applications (Newton, 1996) which may
require more than the basic rectangular block structure of a
single height building, could use this level of roof detail.
Higher detail 3D city models may be necessary to satisfy the
publics inquisitiveness into the finer details of particular
building construction schemes for example.
International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 3W14, La Jolla, CA, 9-11 Nov. 1999
After roof segmentation it is possible to extract height
information for each new segment which could be used to
update current 2D spatial databases with quantitative 3D
information. A degree of semi-automation may be necessary for
this process with the derived roof ridges acting as guidelines
for the user’s own interpretation of the roof structure.
5 SUMMARY AND CONCLUSIONS
This paper has assessed the use of relatively low resolution
LIDAR data for the extraction of roof detail from buildings,
primarily for 3D city modelling. A 2D spatial database of
vector building outlines was used to locate the roof extents.
Two study areas were used to test the LIDAR data, an
industrial area with large simple roofed buildings, and a
residential area with smaller buildings and more complex roof
structures. Through the use of LIDAR elevation, aspect and
slope parameters, attempts were made to extract the main roof
ridge of a building. Results suggest that it is possible to extract
general roof detail using this data, especially for large buildings
with simple roof structures. The aspect parameter performed
best, extracting the largest majority of ridges from the
buildings. Of these extracted ridges using aspect, just over half
can be considered to be exact matches. In the residential areas,
the smaller buildings and more complex roof structures made it
much harder for the LIDAR data to produce meaningful roof
detail.
Other LIDAR data sets with similar or lower resolutions will
most likely suffer the same problems experienced in this
assessment. The vertical and planimetric accuracy levels of the
LIDAR data are not conducive to the extraction of accurately
positioned and located roof detail. The problems with the
LIDAR data were substantial even though the buildings used in
the study were chosen for their relatively simplistic structures.
Despite these limitations, the level of roof detail extracted in
this study is suited to several applications such as visualisation
and spatial database updating. A semi-automated approach to
roof detail extraction may need to be employed for these
applications. The extracted ridges can be used as a guide for
the user to better define the extent and nature of the ridges. To
increase the amount of roof detail that can be extracted from
this LIDAR data, the resolution, or rather the original point
spacing of the LIDAR sensor, should be made more dense.
This will, however, increase the cost of the data and make it
less affordable to most laser scanner users. LIDAR data
accuracy levels need to be improved further, as well as the
relative accuracy levels between LIDAR data and any assisting
data sets such as the 2D vector data set used in this paper. Until
the use of higher density laser scanner instruments becomes
more widespread and the cost of the technology decreases,
relatively low resolution laser scanner data can be used to
extract general roof detail from buildings.
Internatio
AC
We would like to th:
the Ordnance Surve
Survey of Britain |
Thanks to the Env
LIDAR data. Resea
available by the
Nottingham.
Axelsson, P., 1999
algorithms and applic
and Remote Sensing,
Braun, C., Kolbe, T.
Cremers, A.B., Fórst
photogrammetric b
Graphics, 19 (1), pp.
Frère, D., Hendrickx
Gool, L., 1997. On
from aerial images.
objects from aerial
Baltisavias, E. and EF
95.
Grüen, A. and Dan,
for automated buil
extraction of man-m.
Il. (eds) Grüen, A
Birkhauser, Berlin, p
Haala, N. and Brenn
trees in urban
Photogrammetry and
Hug, C. 1997. Ex
airborne laser scanne
made objects from ac
Baltasavias, E. and H
213.
