In: Wagner W., Székely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B 
639 
FUSION OF ALS POINT CLOUD DATA WITH HIGH PRECISION SURVEYING DATA 
A. Wehr 3 ’ *, H. Duzelovic b , Ch. Punz b 
a Institute of Navigation, University of Stuttgart, Breitscheidstr. 2, 70174 Stuttgart, Germany- 
wehr@nav.uni-stuttgart.de 
b rmDATA, Datenverarbeitungs GmbH, Prinz Eugen-Straße 12, A-7400 Oberwart, Austria- 
(duzelovic, punz)@rmdata.at 
KEY WORDS: LID AR, Fusion, Modelling, Algorithms, DEM/DTM, Multisensor 
ABSTRACT: 
In today airborne laser scanning (ALS) extended areas are surveyed with a high point density and with decimetre elevation accuracy 
in a very short time. However, due to the finite sampling process the correct modelling of the surveyed earth surface is difficult, if 
break lines and special topographic features like railway tracks and highways are to be modelled. To improve the ALS derived 
models more and more additional surveying data are used which are measured by e.g. GNSS or tacheometers. These measurements 
have higher accuracy and are sampled in a way that they describe best the features to be modelled. For example break lines are 
described by splines derived from a tacheometric survey. As these supplement data are provided from independent sensors in their 
own coordinate system, all data sets to be fused have to be transformed so that the most accurate model can be computed. This 
means the algorithms must regard data property of the different data sets. In addition the most accurate and precise data set has to be 
used as reference. In this paper algorithms for the fusion of ALS data and additional surveying data obtained from tacheometric and 
DGNSS measurements are presented and discussed based on results of empirical computations on different data sets. The additional 
surveying data consists either of single point measurements or profiles. The presented algorithms are developed under the objective 
to use primarily existing functionalities of a commercial program 
1. INTRODUCTION 
Today airborne laser scanning (ALS) makes possible surveying 
the topography of extended areas with high point density and 
with decimetre elevation accuracy in a very short time. For 
example the ALTM Gemini of the Optech company achieves a 
swath width of 1865 m flying at an altitude of 2000 m and 
sampling data with a point density of about 1.5 m realizing the 
mentioned accuracy. Although this advanced technology 
revolutionized the surveying with regard to the amount of data 
and elevation accuracy, there is still a deficiency in precise 
modelling special topographical features, e.g. break lines, 
highways, railway tracks etc. due to the sampling process. 
However, very often these special topographical elements are 
surveyed by distinct surveying means like GNSS and 
tacheometric measurements, which reach accuracy down to 
millimetres. Therefore, it is obvious to combine these 
complementary measurements for an advanced modelling. In 
addition, it must be regarded that more and more the modelling 
process using ALS data is supported by using information of 
geoinformation systems (GIS). Following the trends in 
generating precise Digital Terrain Models (DTMs) out of ALS 
data makes clear that all available additional information is 
integrated into the modelling process to speed up, to improve 
the robustness of calculations and to increase the precision. 
This paper deals only with the integration of supporting 
surveying measurements obtained by conventional means e.g. 
GNSS real time kinematic and tacheometric measurements. 
These measurements exhibit in general a much lower point 
density but offer a point accuracy which is an order of 
magnitude better compared to ALS data. 
Working with commercial software which derives a DTM out 
off ALS point clouds one very often faces the problem that 
certain structures e.g. break lines or sharp comers etc. are not 
correctly modelled. This problem becomes very obvious, if the 
surveyed surface is modelled by a Triangulated Irregular 
Network (TIN) and the TIN is not modelled accurately to shape 
of the surface. This is especially the case if an unsupervised 
Delauney Triangulation is applied. A typical example 
concerning this case is shown in Figure 1. 
Figure 1. Modelling bad in form to the surface 
(Wehr et al., 2009) showed that the triangulation model can be 
improved, if break line points are regarded in the triangulation 
process (s. Figure 2). In (Wehr et al., 2009) the break line is 
determined out of the point cloud data by a special algorithm. 
However, very often this additional information is already 
available for section of interest from other surveying sensors. 
Fusing data of different sensors the problem arises, that the 
independent data sets are not exactly registered. Therefore, 
supplementary processing steps are required for coregistering. 
The algorithms presented in the following are developed with 
* Corresponding author.