Istanbul 2004
f buildings in
nner data (in
logy.
ght Texture
aser Scanner
nal Airborne
Canada, pp.
3. Automatic
ata for Map
'ogrammetry,
ces, Dresden,
bata, T. and
ngs using an
togrammetry
Iwanami, E.,
r System for
Orthoimage
togrammetry
2, part 4, pp.
‘he -Use of
segmentation
il Archives of
terdam, The
tic structure
ndscape. In:
» on Remote
JRBAN2003,
Comparison
om Airborne
ication in the
Sensing.
h, C., 1999.
for Change
VIIth CIPA
cessed from
te/SFB461/
ty of Object
ng Based on
irchives of
Information
13, pp. 149-
DETAILED BUILDING RECONSTRUCTION FROM AIRBORNE LASER DATA
USING A MOVING SURFACE METHOD
Abdullatif Alharthy!, James Bethel?
'College of Engineering, Umm Al-Qura University, PO Box 555, Makkah, Saudi Arabia
? School of Civil Engineering, Purdue University, 550 Stadium Mall Drive, West Lafayette, IN 47907
alharthy@uqu.edu.sa, bethel@ecn.purdue.edu
KEY WORDS: LIDAR, Application, Detection, Extraction, Building, Reconstruction, Modelling
ABSTRACT:
The increasing demand for a fast, efficient and low cost algorithm for extraction of 3D urban features was the motive behind this
work. In this paper we present a new technique to reconstruct buildings with detailed roofs in urban areas using airborne laser
scanning altimetry data. We have tried to show that dense airborne laser scanning data is sufficient for detailed 3D reconstruction of
urban features such as buildings. This concept is based on local statistical inferences. Least squares moving surface analysis with
variable window sizes and shapes of laser-derived points was the key in determining building roof details. The consistency of the
data with those surfaces determines how they will be modelled. After obtaining the roof facet orientation and approximate location,
the roof boundary will be extracted by intersecting those facets. Consequently a complete wireframe of buildings is constructed.
Results from an actual dense airborne laser data set collected over the Purdue campus are presented in the paper.
1. INTRODUCTION
3D city models are the final outcome of many photogrammetric
applications. In this paper, the approach of reconstructing 3D
building descriptions from LIDAR data is discussed. This
approach can be applied to any DEM data regardless of its
source. However, DEM accuracy plays a major role in defining
the performance of this approach
With the availability of many sources of data such as
conventional imagery, SAR imaging, IFSAR DEMSs, and
LIDAR DEMs, there are many avenues open to derive terrain
and feature data in urban areas. Through much research, it has
been shown that laserscanning data has the potential to support
3D feature extraction, especially if combined with other types
of data such as 2D GIS ground plans (Maas, 1999; Brenner and
Haala, 1999; Weidner and Fórstner, 1995). Despite the fact that
LIDAR data is attractive in terms of cost per high quality data
point, the quantity of the data makes a challenge for storage and
display (Vosselman, 1999). Acquiring 3D object descriptions
from such data is a difficult problem and many approaches have
been tried to solve it. Several of them have succeeded with
some limitations. The principle idea of this research is to detect
and reconstruct buildings form laser altimetry data exclusively.
In earlier work, (Alharthy and Bethel, 2002), we presented an
algorithm to detect building footprints using LIDAR data. The
building detection procedure described includes detecting and
excluding other natural features such as trees. Many
segmentation techniques such as thresholding determined by
histogram analysis, the use of 2D GIS data, and multispectral
inference have been tested together with LIDAR heights to
determine building outlines (Brunn and Weidner, 1997). Using
the second strategy, 2D GIS ground maps give the building
footprints.
After obtaining building outlines itself, the raw data points in
each building polygon will be counted and labeled accordingly.
Processing point sets in each polygon will provide the
necessary characteristics to rebuild the roof surface. In addition
a refinement step to get precise roof details is presented. This
step utilizes the roof planar parameters to partition roof surfaces
into homogenous roof facets. The refinement procedure for roof
segments starts with detecting homogenous roof surfaces and
segmenting them based on their geometrical surface parameters.
Roof outlines are extracted and roof planar facet breaklines are
then determined and refined. After connecting extracted roof
planes, a complete wireframe of processed buildings will be
formed and a 3D view of them will be shown.
2. ESTIMATION OF GEOMETRIC PARAMETERS
FOR MOVING SURFACES
Several algorithms have been suggested to extract roof faces
using range data (Brenner, 2000; Brenner and Haala, 1999;
Brunn and Weidner, 1997; Brunn, 2001; Vosselman, 1999; and
many others). Surface normal segmentation is one of the major
ones. However normal vectors tend to be very noisy due to the
variability in the LIDAR points. In this work, a new technique
to reconstruct buildings is presented. Least squares moving
surface analysis with variable window sizes and shapes of laser-
derived points was the key in determining building roof details.
The consistency of the data with those surfaces determines how
they will be modelled.
2.1 Least squares moving surfaces
A grid with a designed spacing (one meter is used with the test
data here) is overlaid on the irregular LIDAR points in each
building outline as shown in figure 1, where the small green
crosses represent the irregular scattered LIDAR points. Then a
window is moved through the grid cell by cell in both x and y
directions. In each step, the LIDAR points are counted inside
the window and if they exceed a certain limit in number, a
plane fitting procedure is performed. The reweighted least
squares adjustment procedure is used to estimate a unique set of
plane parameters for the fitted points (Mikhail and Ackermann,
1976). In addition to the basic plane parameters, slope in x,
slope in y, and height intercept, the RMSE of the fitted data
over the given window is determined as well, in order to
evaluate the soundness of the recovered parameters. Those
parameters will be recorded at the center cell "pixel" of the
