USING STEREO IMAGES TO DENSIFY LIDAR DATA POINTS AT WHERE NEEDED
Zheng Wang
EarthData Holdings, 7320 Executive Way, Frederick, MD 21704, USA
E-Mail: zwang(@carthdata.com
KEY WORDS: LIDAR, Image, Building, Application, Algorithm
ABSTRACT:
For many applications, dense Lidar data is often needed. For instance, for telecommunication antenna network design, 1 meter
spacing Lidar data would be needed for downtown area of a big city. However, it is expensive to collect high density Lidar data.
Many systems can easily collect 2-3 meter spacing data with one flight line, but for 1 meter spacing data, multiple flight lines
including cross flight lines have to be flown to meet the spacing requirement. In this paper, a novel approach is presented to tackle
the problem. The approach generates 3D points through image matching techniques to densify Lidar points at where denser points
are needed. For example, when denser building points are needed, the approach will first identify all the buildings and then generates
3D points for the buildings. The generated 3D points plus the existing Lidar points will make a denser coverage for the buildings.
The 3D points can be generated at any specified reasonable spacing. The paper first introduces the motivation and need to develop
such an approach, and then describes the concept of the approach and the system design. After those are the experimental results and
conclusions.
1. INTRODUCTION
For many applications, dense Lidar data is often needed. For
instance, in telecommunication antenna network design, 1
meter spacing Lidar data would be needed for downtown
area of a big city. However, it is expensive to collect high
density Lidar data. Many systems can easily collect 2-3
meter spacing data with one flight line, but for 1 meter
spacing data, multiple flight lines including cross flight lines
have to be flown to meet the spacing requirement. So, the
need of minimizing the cost of Lidar data acquisition
motivate people to explore the possibility of using other
alternative(s) to meet the point density requirement. In this
paper, a novel approach of using stereo images to densify the
Lidar data points is presented.
Quite often, when Lidar data is collected, stereo imagery
covering the same area that the Lidar data covers is acquired
as well. We all know that 3D terrain data can be extracted
from a stereo pair of images (left and right images normally
with 60% overlap) either manually or automatically. When it
is done manually, breaklines and mass points are digitized
from a stereo pair of images on a softcopy photogrammetry
system. When it is done automatically, image matching
technique(s) are used to find conjugate points on the stereo
pair of images and then 3D points are formed by the space
intersection process (Moffitt and Mikhail, 1980) for the
conjugate points. Here, conjugate points always come in pair;
one in the left image and the other one in the right image and
a pair of conjugate points represent the same point on the
ground. If a mass of conjugate points generated by the
automated image matching process is dense enough, then the
conjugate points become an additional terrain data source
and can be used as complement to the existing Lidar data
points. When the two data sources are merged together, the
needed point spacing can be met.
Image matching techniques have been used for more than a
decade to generate Digital Elevation Models (DEMs) (Grün
at al., 1995). Image matching works quite effectively and
efficiently for areas where terrain is flat or rolling and ground
surface has rich texture, i.e., the surface texture is not
uniform like sand beach. However, image matching faces
difficulties at where terrain changes abruptly, e.g., steep
mountains and high-rise buildings in cities. The concept of
using image matching to generate DEMs is to find 2D
conjugate points in a stereo pair of images through image
matching technique and then generate 3D ground points by
the space intersection process for the 2D conjugate points.
One of the parameters in the image matching is the Search
Window that determines the searching range of a conjugate
point. The size of the search window is critical to the success
of the image matching process. A too big search window
often leads to a wrong match. On the contrary, a too small
search window easily leads to no match. The search window
should be a function of the terrain elevation change: the
bigger the elevation changes, the bigger the search window
should be. But, when there is no terrain information available
at all for an area, the selection of the search window size can
only be based on any available general knowledge of the
terrain in the area, and therefore the image matching results
are certainly not as reliable as one wants. However, when
there exists available terrain information such as from an
existing Lidar data, the existing Lidar data can make a huge
positive difference on the results of the image matching. The
existing Lidar data can provide accurate and reliable
elevation information about the terrain, which allows an
accurate determination of the searching location and a
relatively small search window, which, in turn, keeps the
image matching time to a minimum.
This paper presents an approach developed at EarthData that
uses stereo images to generate 3D points through image
matching with terrain information from an existing Lidar
data set and then add the generated 3D points to the existing
Lidar data set to meet the Lidar data point density
requirement. For many applications, dense Lidar points are
only needed for buildings, again for example, the
telecommunication antenna network design. Therefore, this
approach was designed to not densify the entire area of a
Lidar data coverage; instead, it only densifies the buildings.
Doing so also keeps the data amount to a minimum.
International ;
In the S
the appr
This se
(LPDS)
Section
results a
the pap
potentia
directioi
2.
An appı
points t
points à
designe
mention
approac
points a
to first]
building
2D conj
the buil
Lidar d
done at
search \
matchin
generate
the spa
building
make a
be gene
produce
assump
Lidar d:
have at
spacing
accurate
Additio:
process
better 1
densifie
output :
stereo it
or finer.
Based
drawing
LPDS.
algorith
The pr
operator
detectec
points.
second
image r
LPDS i:
2D con
2D con
process
process
LPDS i
Lidar bi
also call
