In: Paparoditis N.. Pierrot-Deseilligny M.. Mallet C.. Tournaire O. (Eds). 1APRS. Vol. XXXV111. Part ЗА - Saint-Mandé, France. September 1-3. 2010 
198 
QUALITY MEASURES FOR BUILDING RECONSTRUCTION FROM AIRBORNE 
LASER SCANNER DATA 
Sander Oude Elberink 
Faculty of Geo-information Science and Earth Observation (ITC), University of Twente 
Hengelosestraat 99, P.O. Box 6, 7500 AA Enschede The Netherlands 
oudeelberink@itc.nl 
Commission III, \VG 111/4 
KEY WORDS: buildings, laser scanner data, quality analysis 
ABSTRACT: 
As 3D city models become more detailed and more accurate, it is of importance to correctly judge the quality of these models. The 
problem is that the quality of 3D buildings is composed of multiple indicators that cannot easily fit into a single value. Before 
analysing differences between reference data and 3D models, it is necessary to understand the realisation of the 3D models. The 
following study presents criteria to evaluate the quality of reconstructed building models, from a reconstruction point of view. This 
can be seen as a relative quality check, as no usage has been made of independent reference data. The advantages of the relative 
quality check are that the quality can be predicted for the complete dataset and that it provides information on which object parts are 
less accurately reconstructed than others. In this paper we describe several quality measures on buildings that have been 
reconstructed using airborne laser data. It is shown how the quality measures can be used to gain insight in the quality of the output, 
but also to improve processing steps along the way from laser data to 3D model. If future users have to indicate if the model is 
suitable for certain applications, it is advised to deliver these quality measures together with the reconstructed model. 
1. INTRODUCTION 
Reconstructing buildings in 3D has been a challenging research 
topic for at least ten years, and will be in future as long as 
acquisition systems are improving and model requirements are 
increasing. The tendency is that the reconstructed models 
become more realistic and more detailed. Once the city model 
has been created, it is likely that it is stored at a central location, 
from where multiple users have access to the model. Therefore 
a description on the quality of the model is necessary in order to 
decide if the model can be used for certain applications. Only 
specifying what the Level of Detail (LoD) is does not mean that 
the geometric accuracy of the model has been determined. 
Mostly, general parameters such as minimum footprint size and 
positional accuracy values are mentioned for a certain LoD 
(Kolbe et al., 2005). However, this does not give an insight in 
the quality of the specific building models. 
Figure 1. Various questions concerning the quality of 3D 
buildings. 
The problem is that the quality of 3D buildings is composed of 
multiple indicators that cannot easily fit into a single value. 
Figure 1 shows a number of questions that deal with the 
geometric accuracy of a 3D building. The questions are part of 
the overall question whether the building model is usable for a 
certain application. 
For an absolute accuracy measure, an independent reference 
dataset would be needed in order to check for differences 
between the reconstructed models and the reference dataset. 
These differences contain information on the mean and local 
variation in 3D between reference data set and reconstructed 
model. Reference data can be acquired manually or semi 
automatic, as in (Rottensteiner, 2006). If the reference data is 
considered to be the ground truth, its quality should be better 
than the reconstructed models. The problem is that such a 
detailed reference dataset might not be available (yet) at a large 
scale for detailed models as in Figure 1. When using reference 
data to determine the quality of a certain set of reconstructed 
buildings, it is necessary to correctly analyse differences 
between the two datasets. 
Suppose the building from Figure 1 has been constructed using 
an airborne laser data set. The top ridges have been determined 
by intersection of two roof faces, which are accurately 
described by a plane through planar segments. The top ridges in 
Figure 1 can be determined with a higher accuracy than the 
gutter of the same building. This information is essential when 
analysing differences between the model and reference data of 
either the ridges or the gutters. 
Before checking on reference data, we can predict the quality of 
the reconstructed models using internal quality measures. These 
quality measures can be calculated from the input data, and can 
therefore not be seen as independent. However, it is a measure 
for the expected quality. The advantages are that the quality can 
be predicted for the complete dataset and that it provides 
information on which object parts are less accurate than others. 
When such a quality description is directly attached to the 
reconstructed building models, it will be possible to perform a 
stochastically correct quality check using reference data once 
available. Especially in the phase of creating city models, 
instead of updating these models, it is important to describe the 
expected quality using internal measures.