ANALYSIS OF TIN-STRUCTURE PARAMETER SPACES IN AIRBORNE 
LASER SCANNER DATA FOR 3-D BUILDING MODEL GENERATION 
A. D. Hofmann 
Institute of Photogrammetry and Remote Sensing 
Dresden University of Technolog 
Mommsenstr. 13 
D-01062 Dresden, Germany 
Alexandra.Hofmann@mailbox.tu-dresden.de 
http://www.tu-dresden.de/fghgipf/forschung/BuildingDetectionAndModelling/default.htm 
Commission III / WG3 
KEY WORDS: LIDAR, Triangulation, Cluster Analysis, Building model reconstruction, Geometric accuracy 
ABSTRACT: 
This paper presents a data-driven method for automatic building reconstruction from raw airborne laser scanner data. The method 
utilises a TIN-structure that is calculated into the point cloud. The parameters of every TIN-mesh are mapped into a 3D triangle- 
mesh parameter space, which is then analysed using a cluster analysis technique. By analysing the clusters, significant roof planes 
are derived from the parameter space while taking common knowledge of roofs into account. However, no prior knowledge of the 
roof as such as the number of roof faces is required. Analysing the intersected roof faces, the roof outlines are determined and the 
ground plan is derived. 
The derived building models were evaluated for their correctness and geometric accuracy. Well-defined building roof planes can be 
extracted and reconstructed successfully, while disturbances such as dorms on buildings or geometric discrepancies in laser scanner 
data strip overlaps may significantly reduce the applicability of the technique. 
1 INTRODUCTION 
Airborne laser scanner data provides reliable and dense x,y,z- 
coordinates of a surface. With this data, exact information of 
roof shapes can be derived. Different methods have been 
published that use laser scanner data to obtain building 
parameters. Some of them are based purely on laser scanner 
point clouds, while others integrate additional information. The 
former method is more adaptable. Up to now there are two 
types of approaches that run without additional information and 
that derive the full 3D description of a building. The first kind, 
a model-driven approach, is parameter-based and was 
developed by [Maas 1999], who uses invariant moments. The 
method is, as described in the article, limited to simple ground 
plans, while it can handle relatively complex roof constructions. 
The second type is data-driven. [Rottensteiner 2002] and 
  
s 
Figure 1-1. Building model reconstructed from airborne laser 
scanner data, orthophoto 
This paper will discuss a further procedure for building model 
[Elaksher 2002] developed methods that group pixels in reconstruction. It works automatically and is not bound to 
rasterised data that fit in a plane. The results are promising; simple roof types or to a certain point density. The method uses 
however, interpolating the laser scanner data to a raster can hide a TIN-structure that is calculated into a laser scanner point 
aluable information. [Gorte 2002] and [Lee 2001] avoid this cloud that contains a building. The position of each triangle in 
by applying a region-growing algorithm to a TIN-structure in space is expressed in spherical coordinates that are displayed in 
raw laser scanner data. The difficulty here is to decide which à Cartesian coordinate system. These coordinates are taken as 
neighbouring segments should be merged. [Vosselmann 1999] parameters that, consequently, define a 3D parameter space. 
analysed laser scanner points of high density data and The distribution of points in parameter space (representing 
transformed them into a parameter space. In combination with a triangles in object space) shows a structure that is analysed with 
segmentation procedure the approach has encouraging results. a cluster analysis technique. The clusters contain those points 
of the parameter point cloud that belong to triangles of roof 
faces. A plane is interpolated into the laser scanner points of 
the triangles of each roof face. The individual planes of each 
   
  
   
    
    
    
    
     
  
  
  
   
  
  
  
  
  
  
   
   
    
   
     
     
  
  
      
    
   
    
   
    
   
   
   
    
   
    
    
   
    
   
   
Internatio 
laser scar 
order to 
assessed f 
At the sta 
data. Th 
data sets. 
1.5m Thi 
such as d 
point coo 
and in z 2 
region an 
rasterised 
several si 
rather con 
For the : 
building 1 
extracted 
[Hofmanr 
such lase 
clouds co 
gable an 
combinati 
  
The basic 
TIN in the 
roof face) 
space. C 
therefore 
modelling 
the laser | 
describe t 
the build 
paragraph 
In many 
building c 
points of 
inconsiste 
strip’s poi 
triangulat 
Figure 2- 
analyses,