. Istanbul 2004
AUTOMATIC BUILDING EXTRACTION AND 3-D CITY MODELING FROM LIDAR
DATA BASED ON HOUGH TRANSFORMATION
KAZUO ODA TADASHI TAKANO TAKESHI DOIHARA RYOSUKE SHIBASAKI**
ASIA AIR SURVEY CO, LTD.
*THE UNIVERSITY OF TOKYO
Commission III, PS WG III/3
KEY WORDS: LIDAR, Modeling, City, Graphics, Edge, Visibility Analysis
ABSTRACT:
This paper proposes a new method of LIDAR data processing to automatically extract building and 3-D city modeling. This method
exploits image processing technique including Hough transformation. The process of building extraction starts with DSM
segmentation according to elevation values. Among these fragments, extracting process limits candidates of building fragments to
those that have elevation higher than a specified threshold value and the DEM at the same location. Hough transformation is
executed to boundary pixels of each candidate of building fragment and then fit to linear segments. Each fragment is polygonized
with these line segments. After the extraction, 3-D-building model is created with these polygons such that each polygon has vertical
wall from the top of building to the ground (DEM). Texture images are pasted onto each building, including walls. The most
appropriate texture image is selected from aerial images according to geometry between building faces and external parameters of
the photos. The method has been tested with LIDAR data of Shinjuku and Ginza area in Tokyo.
1. INTRODUCTION
Advances of information technology are changing the concept
of map form simple 2-D space expression into 3-D real world
model constructed in cyberspace. There are two main problems
for 3-D model applications. The first one is how to express and
define 3-D model for general use or specific applications. The
other one concerns methods to build up 3-D model at a
reasonable cost performance. This paper focuses on the method
to build up 3-D city model with LIDAR data.
The most typical commercialized method of 3-D city modelling
with LIDAR data simply stands up buildings and houses in 2-D
map with height values of footprints that fall into building
polygons. This method, however, cannot express details of
building shape, since building polygons in 2-D map simply
trace outside boundaries of buildings. Another type of 3-D city
model expression uses original LIDAR DSM. This type of 3-D
model can express the details of city, but the amount of data
tends to be too large and hard to manage or view in PC without
special devices.
This paper presents an automated method for 3-D city model
production with LIDAR data and aerial photo images, which
can be applied to production of 3-D map for infrastructure. First,
LIDAR DSM image is divided into fragments where DSM
pixels have almost the same elevation value. Hough
transformation extract straight lines and curves of building
boundaries and constructs them into 3-D shapes. Extracted
polygons are shaped into prisms with planar walls and aerial
photo images are pasted onto each faces of building model as
texture images according to geometrical information. Thus
building models are more realistic with detail of shapes and
texture images.
This paper first overviews the strategy of building extraction
and construction and introduces building extraction with Hough
transformation. Some experimental results with tests are
presented in the following section with LIDAR data of Shinjuku
and Ginza area in Tokyo, Japan.
2. THE STRATEGY OF THE METHOD
The strategy of the proposed algorithm consists of two parts.
The first part is building extraction where building polygons are
extracted from DSM and aerial photos. The second part is 3-D
modelling where 3-D-building model is created with these
polygons such that each polygon have vertical wall from the top
of building to the ground.
2.1 Building Extraction
The schema of the algorithm is shown in Figure 1.This method
treats DSM data as “image”, and exploit image processing
technique including Hough transformation. As pre-processing,
LIDAR data are converted in raster format and DEM is
extracted by minimum filtering. Details of the image processing
sequence are as follows.
2.1.1 DSM Segmentation: The process of building
extraction starts with segmentation, where raster DSM is
divided into fragments where DSM pixels have almost the same
elevation value. Among these fragments, extracting process
limits candidates of building fragments to those that have higher
elevation more than a specified threshold value and the DEM at
the same location. Jaggies around DSM masks are cleaned up
with morphological dilation and erosion. A DSM fragment
extracted here is called “DSM masks”. To avoid extraction of
trees, cars and other objects, city block polygons in existing
digital map data can be applied to omit DSM masks outside the
polygons.
2.1.2 Edge Extraction:
Edge Extraction of the DSM image is executed with some edge
operator, such as Canny operator or other zero-crossing
operators. If DSM resolution is too poor to extract edge lines,
aerial photo images can be used alternatively. In resultant edge
image, each edge should be thinned into 1-pixel width edge
lines.