ACCURACY ASSESSMENT OF BUILDING MODELS CREATED FROM LASER 
SCANNING DATA 
A. Borkowski, G. Józków 
Institute of Geodesy and Geoinformatics, Wroclaw University of Environmental and Life Sciences, Grunwaldzka 53, 
50-357 Wroclaw, Poland - (borkowski, jozkow)@igig.up.wroc.pl 
KEY WORDS: Architecture, Modelling, LIDAR, TLS, Accuracy, Three-dimensional 
ABSTRACT: 
Recently, it can be observed a growing interest in 3D building or city models created from laser scanning data. These models are 
used in many areas of interest. In this work the accuracy assessment of 3D buildings models created from airborne and terrestrial 
laser scanning data was carried out. TLS data for modelling were acquired with average point spacing about 0.02m. In order to model 
invisible from the ground building elements such as roofs, the LIDAR data was used with density of about 12pts./m°. TLS and 
LIDAR data were acquired in the same coordinates system. For textures mapping a set of 10Mpix digital images using SLR camera 
Canon 40D were taken. Modelling was performed using both commercial software Leica Cyclone, as well as original software and 
free software. The accuracy of models is affected by many factors, but the impact of all factors leading to errors of the final models 
were analysed together. The accuracy of models was assessed by comparing the coordinates of characteristic points of the models and 
the corresponding coordinates of these points measured on real buildings. The accuracy of points measured on the real building was 
not affected by model errors, therefore they can be treated as error-free reference points. On site reference measurements were 
performed using Leica TCR407Power reflectorless total station. Obtained inaccuracies of the reconstructed models were as follow: 
0.12m horizontal, 0.13m vertical, 0.18m three-dimensional. 
1. INTRODUCTION 
Technological advances in airborne laser scanning on the one 
hand and the greater availability of laser scanning techniques on 
the other hand cause in recent years rising interest in building 
3D models created based on laser scanning data. Building 3D 
models are increasingly being developed for the purpose of 
promoting city or region, risk management, administration or 
spatial analysis such as analysis of the insolation, the 
propagation of noise, pollution propagation, etc. 
3D modelling of buildings from laser scanning data is also 
a current research problem, whichever is here mainly for the 
development of methods and algorithms to automate the 
modelling process, based primarily on airborne laser scanning 
data. Such modelling is a complex process in which a number 
of problems have to be solved, from the identification of the 
building in the data set to the reconstruction the vector model. 
Review and discuss the issues in this area can be found in the 
current works (Vosselman and Maas, 2010; Awrangjeb et al., 
2010, Keller and Borkowski, 2011). 
Cloud of points recorded by terrestrial laser scanner is in itself a 
good representation of a building model due to the high 
scanning resolution. However due to the size of data sets such 
model is not very practical in use. Therefore there is a need to 
create a geometric 3D model. Automatic modelling in the case 
of terrestrial scanning data refers rather to the modelling of 
building facades (Pu and Vosselman, 2006; Boulaassal et al., 
2010). 
Terrestrial laser scanning data are generally not sufficient to 
model the geometry of the roof. This information may be 
obtained from the airborne laser scanning data, but it contains 
of the fragmentary information on the facades of buildings. 
Complete data for modelling the geometric construction is 
obtained when the data sets of airborne and terrestrial laser 
scanning are combined. Modelling based on the combined data 
sets do not differ in principle from modelling based on airborne 
or terrestrial scanning data. Some problems may arise at the 
stage of sets co-registration of both data sets due to the different 
precision of the sensors, especially different horizontal accuracy 
for points collected by these sensors. 
In this paper an accuracy assessment of the 3D model on the 
example of several heritage buildings was performed. 
Geometric models were built on the basis of the combined point 
cloud of airborne and terrestrial laser scanning. 
The problem of assessing the accuracy of 3D modelling has 
already appeared in the literature. The work (Akca et al., 2010) 
proposed evaluation of the internal accuracy of model on the 
basis of three-dimensional surfaces matching. Vosselman 
(2008) uses the roofs of buildings modelled on the basis of 
points of two adjacent scan strips to evaluate the horizontal 
accuracy of airborne laser scanning data. Indirectly, this is also 
an internal accuracy assessment of modelling the roofs of 
buildings. The work (Oude Elberink and Vosselman, 2011) 
presents an comprehensive analysis of errors affecting the 
accuracy of the modelling and accuracy assessment of 
modelling the various elements of the building (mainly the 
roofs). The proposed method of assessing the accuracy is based 
on point clouds analysis - accuracy parameters resulting from 
the comparison of created model and laser scanning data. 
In this work, the opposite approach has been presented. The 
assessment of accuracy was based on a comparison of selected 
elements of the model with the results of measurement of these 
elements made directly on site. This assessment is based on an 
analysis of differences between two sets of data and is 
postulated in the paper (Oude Elberink and Vosselman, 2011). 
The assumption for this analysis is the availability of accurate 
reference data. In this work the role of precise reference data 
fulfil data from tachymetric measurement of object. To proceed 
the accuracy evaluation of the model, a brief description of the