125 
BUILDING EXTRACTION FROM ALS DATA BASED ON REGULAR AND 
IRREGULAR TESSELLATIONS. 
Natalia Borowiec 
AGH University of Science and Technology in Krakow, Department of Geoinformation, Photogrammetry and 
Remote Sensing of Environment - nboro@agh.edu.pl 
Youth Forum 
KEY WORDS: LIDAR, Extraction, Detection, Building, TIN, GRID 
ABSTRACT: 
For the past dozen years, data which enable the use of Digital Terrain Model as well as Digital Surface Model have been widely 
used. One of the modem techniques of gaining information about terrain is the Airborne Laser Scanner. We can obtain the spatial 
coordinate points using the laser scanner, which creates a set of large density points commonly called the "cloud of points". These 
models are reliable and accurately reflect the reality surrounding us. For pictorial character purposes, this kind of data is sufficient. 
However in the case of modeling buildings, structure and engineering objects, it is necessary to process data which includes: 
filtration (removing noise and excessive data) in order to detect points and essential lines needed for spatial object description and to 
organize a vector description of the modeled objects.This paper presents a new method of extraction. It also presents a preliminary 
roof edge extraction based solely on laser data, without the help of additional information. The idea of the proposed method is based 
on regular and irregular tessellations, made utilizing the last echo. 
1. INTRODUCTION 
Owing to GIS development a strong demand for terrain models 
(DTM) and models of surface determined by terrain overlay 
(DSM) is observed. These models are used in various types of 
GIS analysis, i.e. municipal and rural area’s changes planning, 
definite spatial occurrence’s effects forecasting (i.e. propagation 
of noise or air pollution). As a consequence of high demand for 
3D models, the automatic extraction of natural and artificial 
elements placed on the ground surface is a subject of many 
publications. 
One of methods, in which DTM and DSM is acquired is 
stereodigitalisation of photogrammetrial models. However, the 
photography’s resolution not always enabled having a high 
precision object models. Therefore, a field for a new method 
progressing turned out, allowing acquisition of spatial models; 
this is the LIDAR technology. During last two decades, a broad 
application and use of airborne laser scanning (ALS) in process 
of 3D objects modelling is emphasized in many publications. 
The demand for 3D buildings models (DBM) is growing 
particularly fast. Up to now the buildings are saved in GIS 
systems as a 2D objects, which strongly limits possibilities of 
spatial analysis. It is worth to mention, that single buildings 
shape modelling in three dimensions is a high complexity rank 
task, much more composite than to build a generalized DTM 
type areas for the geographical data systems purpose. 
1.1 Related work 
ALS’ result is “points cloud”, which is composed of points 
located either in a terrain or on overlay elements. That is why, 
the first step in the process of laser data processing is a 
distribution of points to these, which form the terrain and these 
which belong to terrain overlay objects. Many solutions of 
grouping points to both mentioned classes can be found in 
publications. The 3D objects extraction is strictly linked to 
points grouping. 2 approaches can be distinguished among 
methods of building detection: 
a) supporting of the ALS of additional data i.e. airborne and 
satellite photos or registry maps, 
b) exploiting only ALS data. 
Taking into consideration the difficulty of solution, the second 
approach is more demanding and work consuming than the first 
one. However, despite arising difficulties, using ALS data only 
to object detection is both a scientific and a practical challenge. 
In this approach there is no need of making use of additional 
data, which are not always of appropriate quality, and often 
require high financial costs. 
Objects detection in the approach based on ALS data only is 
divided into two subgroups. The division complies with the 
character and lidar data layout. First solution is to interpolate a 
regular grid from irregular cloud of points. Further process is 
done through commonly known digital image analysis, i.e. 
filters, segmentations (Maas, 1999; Tovari and Vogtle, 2004). 
However, second solution concentrates on the analysis of 
original cloud of points (Morgan, Habib; 2001; Schwalbe, 
2003). Operating on irregular data of high capacity is less 
recognizable, and one of the barriers is the data capacity. In 
sum: grid data analysis are easier but less precise, whereas 
dispersed data analysis are harder to perform but they 
potentially should bring better results. 
1.2 Position of proposed approach 
The method presented in the article is based on emerging 
building data from lidar data only. Whereas, as a whole it does 
not belong to none of above described solutions taking into 
consideration the data character. Its particular attribute is a 
stage configuration, on particular stages grid or - originally - 
dispersed data are used.