ISPRS Commission III, Vol.34, Part 3A „Photogrammetric Computer Vision“, Graz, 2002 
  
mean is taken. If a string has a numeric score at one end 
and a *no score" at the other end, the numeric value is 
used. If a string has a “no score” at both ends, then zero is 
used. In this way an overall score is obtained for each 
surface. Reinterpreted the score indicates if a surface is 
“terrain/unclassified” (values close to 0) or “object” (values 
close to 1). Seen in Figure 4(e), the algorithm has identified 
the roofs as “object” regions (here shown as dark gray) 
that should pose no problems for filtering. This leaves the 
lighter regions (values close to 0) that represent areas that 
maybe terrain or “difficult to classify”. In the first instance 
“terrain” and “difficult to classify” regions can be 
discriminated by an examination of surface areas. The 
larger the surface area the more likely that the surface is 
terrain. Once terrain (large areas) areas have been 
determined in this way, then, remaining terrain/unclassified 
regions can be tested against the already determined terrain 
surfaces. If a surface is below already determined terrain 
then very likely it too is terrain. Further to this low points 
in the point-cloud can be determined. If these points 
coincide with surfaces determined as terrain, then this 
would serve as confirmation. 
2 “ 
6. Once surfaces have been coded as “terrain”, “object” or 
“difficult to classify” then every point in the point-cloud 
also has to be coded. This is done by assigning to every 
point the score of its corresponding surface. 
Once surfaces have been coded the search for ambiguities also 
becomes possible by examination of the string scores. For 
example a terrain surface should not have strings whose scores 
are close to 1. In here lies the strength of the procedure in that 
the local context (string scores) are used to obtain a global 
context (surface scores) for the landscape. The above procedure 
is still very much under development but is shown here to 
demonstrate a possible way for identifying regions that might be 
difficult to classify. Moreover, the detection of contradictions 
and the determination of the influence of gaps on strings should 
provide more realistic results. Furthermore, the manner in which 
strings were generated is not yet ideal because it is based solely 
on height difference. Future, implementations will use the area 
of the different segments and second returns to strengthen the 
classification process. 
| oan | 
rA 
| dh 
rg ol Jo 
No score 
Figure 5 Assigning scores to string ends (strings separated 
in height). 
5 CONCLUSION 
In this paper it has been proposed that in a preprocessing step a 
point cloud should be segmented and that the results of this 
should drive the filtering. Further still it is argued that this 
preprocessing step is essential if external information is to be 
used in a filtering procedure, to clarify or validate the 
classification of regions that cannot be classified on positional 
information alone. Some of the characteristics of the landscape 
and data that lead to a difficulty to classify laser point-clouds 
have been highlighted. These characteristics are then used to 
search for regions in the point cloud that cannot be classified 
with certainty. An initial attempt at developing an application to 
segment such regions has been demonstrated. However, this 
application still requires much improvement and there are also 
other aspects such as identifying contradictions, and data gaps, 
data resolution, etc., that will have to be treated. 
REFERENCES 
Axelsson P., 1999: “Processing of laser scanner data - 
algorithms and applications”. ISPRS JPRS, 54 (1999). pp. 138 - 
147. 
Ackemann F., 1999: “Airborne laser scanning — present status 
and future expectations”. ISPRS JPRS, 54 (1999). pp. 64 -67. 
Elmqvist, M, 2001: “Ground Estimation of Laser Radar Data 
using Active Shape Models”. Paper presented at the OEEPE 
workshop on airborne laserscanning and interferometric SAR 
for detailed digital elevation models 1-3 March 2001, paper 5 (8 
pages). Royal Institute of Technology Department of Geodesy 
and Photogrammetry 100 Stockholm, Sweden. 
Gooch, M., Chandler, J. 1999; “Failure Prediction in 
Automatically Generated Digital Elevation Models". 4th 
International Conference on GeoComputation, Virginia, USA, 
1999, CDRom. 
Haugerud R.A., Harding D.J., 2001: “Some algorithms for 
virtual deforestation (VDF) of LIDAR topographic survey data”. 
IAPRS, Vol. 34-3/W4. pp. 211-218. 
Lee, I. and Schenk T., 2001: “3D perceptual organization of 
laser altimetry data”. IAPRS, Vol. 34-3/W4. pp. 57 - 65. 
Kraus, K., Pfeifer, N., 1998: “Determination of terrain models in 
wooded areas with airborne laser scanner data. ISPRS JPRS. 
Vol. 53, pp. 193-203. 
Petzold B.; Reiss P.; Stossel W.; 1999: “Laser scanning — 
surveying and mapping agencies are using a new technique for 
the derivation of digital terrain models.” ISPRS JPRS. Vol. 54 
(1999). pp. 95 — 104. 
Vosselman, G., 2000: “Slope based filtering of laser altimetry 
data”. IAPRS, Vol. 33/B3, pages 935-942. 
A- 335