In: Paparoditis N., Pierrot-Deseilligny M.. Mallet C.. Tournaire O. (Eds), 1APRS. Vol. XXXVIII. Part ЗА - Saint-Mandé, France. September 1-3, 2010 
202 
containing these segments can either be shown to the users for 
manual interpretation or can serve as input for an alternative 
reconstruction approach. The latter was the case in Figure 8 
where constraints on the roof inclination were changed from 
‘equal for all hip roof faces’ (top row) to ‘equal for two 
opposite roof faces’ (bottom row). 
at situations where water is standing in a comer of dormer. In 
this case, the model is correct; however the quality measure 
indicates a high value. Knowledge on the expectation of the 
quality of the model at this point as explained in section 3, 
helps interpreting and accepting the larger distance between 
data and model. 
Figure 8 Top row: hip roof reconstructed with equal inclination 
angles. Right: laser point residuals superimposed. 
Bottom row: hip roof reconstructed with two 
different inclination angles. Right: as a result, the 
improvement on the laser point residuals is directly 
visible. Map polygon is shown as yellow polygon. 
Figure 10 Nearest distance between 3D model points and laser 
points, coloured by residual value (<20 cm is green, 
<50 cm is yellow, > 50 cm is red). Right: Roof laser 
segments projected on wireframe. 
Complete segments that have not been used for building 
reconstruction are also mentioned in (Oude Elberink and 
Vosselman. 2009) where the reasons were discussed. The 
recording of these segments is enough to have a quality 
measure of the reconstructed buildings, as it directly contains 
information on the completeness of the reconstructed buildings. 
The disadvantage of calculating the perpendicular distance 
between laser points and model faces is that it can be 
misleading in the sense that most of the laser points show a 
small residual. This is especially the case for data driven 
approaches that fit each individual roof face through a roof 
segment. It does not show the quality of the location of the 
edges of the roofs. Another quality measure is given by 
calculating the distance between 3D comer points and the 
nearest laser point, as visualised in Figure 9. Although this is 
not an independent check either, it is of added value to the 
previous height check, because it holds information on how 
assumptions and constraints on the edge locations fit to the 
data. 
Figure 9 The distance between a model point and nearest laser 
point (indicated by red arrows) as a quality measure. 
In Figure 10 two situations are shown in grey circles, where the 
corner points of building models have a large distance to the 
nearest laser point. A gable roof, which is reconstructed using a 
model driven approach, is shown in the front part of the scene. 
The constraint that the gable roof is symmetric is a basic 
assumption from the model driven approach. However, using 
the distance to the nearest laser point in the corresponding roof 
segment, it is detected that one gutter has incorrectly been 
reconstructed. Rejecting the symmetric constraint is the 
solution for this case. The other situation shows a dormer which 
coiner points do not have nearby laser points. This often occurs 
5. QUALITY MEASURES AS OUTPUT ATTACHED TO 
3D MODELS 
So far, situations have been shown that show a probable cause 
of which some have a relative simple solution. These solutions 
can be implemented such that the best alternative is chosen 
automatically for each of the situations. In practice however, it 
is still advisable to deliver these quality measures together with 
the reconstructed model. Future users can then indicate if the 
model is suitable for certain applications. 
Figure 11 Reconstructed model (upper left) is only part of the 
output; nearest distance between comer points and 
laser points (upper right), orthogonal distance 
between laser points and roof face (lower left) and 
segments not used by the reconstruction algorithm 
(lower right). 
An example is shown in Figure 11 where the reconstructed 
model (upper left) is only part of the output of the building 
reconstruction algorithm, as three quality measures are attached