EDED 
on, 1 meter 
Lidar data. 
flight lines 
ed to tackle 
enser points 
^n generates 
e buildings. 
1 to develop 
| results and 
and ground 
ture is not 
ching faces 
eg. steep 
concept of 
to find 2D 
ough image 
id points by 
gate points. 
; the Search 
a conjugate 
! the success 
rch window 
a too small 
rch window 
change: the 
rch window 
on available 
low size can 
edge of the 
hing results 
vever, when 
as from an 
nake a huge 
atching. The 
ind reliable 
1 allows an 
ition and a 
1 keeps the 
wthData that 
ough image 
isting Lidar 
the existing 
pint density 
ar points are 
ample, the 
erefore, this 
re area of a 
ne buildings. 
im. 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
  
In the Section 2, the paper introduces the general concept of 
the approach and some basic considerations in the approach. 
This section also shows a Lidar Point Densification System 
(LPDS) developed based on the approach. Then in the 
Section 3, the paper presents the experimental data and 
results and gives an analysis of the results. In the last section, 
the paper discusses the feasibility of the approach, the 
potential applications of the densified Lidar data, and gives 
direction of future works at the end. 
2. THE APPROACH 
An approach has been developed at EarthData to generate 3D 
points through image matching techniques to densify Lidar 
points at where denser points are needed. The approach is 
designed to densify only buildings for practical reasons 
mentioned in the Introduction section. The concept of the 
approach is described as follows. When denser building 
points are needed for an existing Lidar data, the approach is 
to firstly identify all the buildings by a semi-automated 
building extraction process (Wang, 2000). Then, it generates 
2D conjugate points through image matching with the help of 
the building Lidar data. As explained above, the existing 
Lidar data allows the searching of a conjugate point to be 
done at an accurate searching location and with a minimum 
search window, which therefore leads to a minimum image 
matching time. After the image matching process, the 
generated 2D conjugate points are converted to 3D points by 
the space intersection process. Then, the generated 3D 
building points and the existing Lidar points are merged to 
make a denser coverage for the buildings. The 3D points can 
be generated at any reasonably specified spacing. In order to 
produce reasonable results, the approach has certain 
assumptions on the input Lidar data and imagery: the input 
Lidar data should have 2-3 meter or better point spacing and 
have at least 20cm elevation accuracy. If the Lidar point 
spacing is too coarse, it may not be able to provide an 
accurate searching location for the conjugate points. 
Additionally, the input imagery used in the image matching 
process must have a ground pixel resolution that is equal or 
better than the specified output point spacing for the 
densified data, i.e., if 1 meter point spacing is wanted for the 
output 3D points, then the ground pixel resolution of the 
stereo images used in the matching process has to be 1 meter 
or finer. 
Based on the approach, the LPDS was developed. The 
drawing in Figure 1 shows the processes and data flow of the 
LPDS. The first process of LPDS is Building Extraction. The 
algorithms for this process were developed several years ago. 
The process is semi-automated; it mainly requires an 
operator to delete those non-building features that are 
detected as buildings. This process generates Lidar building 
points. Each building has its own set of Lidar points. The 
second process is the generation of conjugate points through 
image matching. The image matching technique used in the 
LPDS is Cross-Correlation. This process produces a mass of 
2D conjugate points for each and every building. Then the 
2D conjugate points are converted to 3D points in the third 
process that is the space intersection. The output of the third 
process is the 3D building points. The last process of the 
LPDS is to merge the 3D building points with the existing 
Lidar building points to form the densified Lidar points, or 
also called in this paper the densified building points. 
Lidar Data ) 
| Building Extraction | 
  
| Lidar Building Points 
  
    
   
  
Stereo Images 
with Exterior and 
Interior Orientations 
Generation of Conjugate Points 
Through Image Matching 
  
2D Conjugate Points 
  
  
Conversion of 2D Conjugate 
Points to 3D Building Points by 
Space Intersection 
  
  
  
3D Building Points 
  
  
Merge of 3D Building Points 
and Lidar Building Points 
  
  
  
Densified Building Points 
Figure 1. The processes and data flow of the LPDS. 
3. THE EXPERIMENTS AND RESULTS 
3.1 The Experimental Data 
  
Figure 2. The experimental area and buildings. 
The experimental data was a Lidar data with 0.5 meter 
spacing with an average 15 cm vertical accuracy. Then, this 
original 0.5 meter spacing Lidar data was thinned to generate 
a Lidar data set of 3 meter spacing and the generated Lidar 
data set was used in the experiments. Having the original 0.5 
meter spacing Lidar data available allows a quality check and 
analysis on the densified building points. The imagery used 
in the experiments had a range of ground pixel resolution 
from 0.1 to 0.5 meter. The experimental area covered three 
    
  
  
)