EXTRACTION OF 3D STRAIGHT LINES USING LIDAR DATA AND AERIAL IMAGES 
A. Miraliakbari, M. Hahn, H. Arefi, J. Engels 
Dept, of Geomatics, Computer Sciences and mathematics, Stuttgart University of Applied Sciences, Stuttgart, Germany 
(miapgl 10, michael.hahn, hossein.arefi, johannes.engels)@hft-stuttgart.de 
Commission IV, WG 9 
KEY WORDS: Building, Edge, Extraction, Feature, LIDAR, Reconstruction 
ABSTRACT: 
Light Detection and Ranging (LIDAR) is a technology used for collecting topographic data. Nowadays there are diverse application 
areas of LIDAR data that include DTM generation and building extraction. 3D representations of buildings extracted from LIDAR 
data are often quite inaccurate regarding their location of the building edges. High resolution aerial images have the potential to 
accurately extract 3D straight lines of the building edges. Thus using aerial images to extract building edges accurately is an option 
in particular if aerial images are available in addition to LIDAR data. In this paper we present a study on straight line parameter 
estimation based on the line parameterisation proposed by Schenk (2004). We assume that an aerial triangulation is carried out in 
advance, hence the exterior orientation parameters of all images are known. The LIDAR data are used to get approximate values of 
straight line parameters and to simplify the line correspondence problem in the aerial images. The straight lines are extracted from 
the aerial images and the 3D line parameters are estimated. For extraction of straight lines in the aerial images three techniques are 
used. Two of them are dealing with semi-automatic straight line extraction and third one focuses on automatic line extraction using 
the Hough transform. Quality of the results is investigated using different number of images and different number of points 
measured on the building edges 
1. INTRODUCTION 
Light Detection and Ranging (LIDAR) is a technology utilized 
in airborne laser scanning systems for collecting 3D point 
clouds which implicitly represent topographic data. The LIDAR 
data is an excellent source for generating digital surface models 
and digital terrain models. In this processes the irregular point 
clouds are filtered and often interpolated to a regular grid in 
which the height is given as a function of the rasterised 2D 
location. 
An important application in which LIDAR and aerial images 
are used is city modelling. Straight lines, which are in the focus 
of our research, are important features in this context especially 
if man-made objects have to be reconstructed. 
Schenk (2004) discussed the use of straight lines in aerial 
triangulation. He introduced a four parameter representation of 
3D straight lines, where he utilized azimuth and zenith of the 
straight line and the intersection of the straight line with the XY 
plane of a rotated coordinate system. 
In related research, Abeyratne (2007) used simulated data for 
image resection and EOP estimation. He compared methods of 
six parameter and four parameter representation of straight lines 
with traditional point based photogrammetry. 
Lines in the high resolution aerial images do not have this kind 
of deficit thus it would be near at hand to use images for 
extracting 3D lines. But extracting 3D straight lines from 
overlapping aerial images is much more complicated than 
extracting 3D lines directly from the LIDAR data. 
Figure 1. Magnified area of superimposing the LIDAR data 
with ortho photo 
If both kinds of data are available the simultaneous use of 
LIDAR and image data will reduce the complexity 3D line 
reconstruction. This can be considered as an approach to 
improve the accuracy of LIDAR based 3D line reconstruction 
by using the aerial images. 
In the following we give some more theoretical background, 
point out our methodology, discuss the achieved results and 
draw some conclusions. 
A starting point of our research is shown in figure 1. The DSM 
generated from LIDAR data is overlaid with an orthophoto of 2. THEORETICAL BACKGROUND 
the corresponding area. The boundary line of the building roof 
looks rather fuzzy thus no nice straight line can be recognised. Points are the basic elements in most algorithms of 
This is mainly an impact of the interpolation in the LIDAR data photogrammetry. Influenced by developments in computer 
thus it is to be expected that the roof boundary can not be vision however there has been efforts to develop feature based 
extracted very accurately from the LIDAR data. method also.