features in each sub- 
| are merged together 
ar features along the 
nts of linear features 
ndary, see red points 
of upper and lower 
eration. In order to 
poundary, all linear 
al direction based on 
step is performed in 
> bridge is rotated to 
boundaries can be 
rdinate. For straight 
tion is used to fitting 
boundary, 2" order 
polynomial function 
undary can be then 
| computed via the 
ire 10. 
  
i 
puted via 2° order 
the Hough linear 
| 
R intensity and aerial 
can be transformed 
ity image as well as 
nd lower boundaries 
oarse boundary from 
smooth boundaries, 
| between the smooth 
blue) extracted from 
a precise DBM. In 
1 boundary points are 
our points’ elevation 
ith the DTM derived 
  
7.588 
  
7.5875 
1.8545 1.8546 1.8547 1.8548 1.8549 1.855 
  
        
  
L 
1.8548 1.8549 1.855 
  
  
   
  
1.8545 1.8546 1.8547 
x10 
(b) 
Figure 11. Smooth boundary points computed via 2" order 
polynomial function (blue) according to the Hough linear 
feature endpoints (red) 
4. RESULTS AND CONCULUSION 
A comparison between bridge ROI DSM derived from LiDAR 
data and DSM based on the precise DBM is given in Figure 12 
and Figure 13 for the test bridge 1and 2, respectively. 
  
(b) 
Figure 12. Bridge ROI DSM derived from LiDAR data (a) 
and DSM based on the precise DBM (b) (bridge 1) 
33 
  
(b) 
Figure 13. Bridge ROI DSM derived from LiDAR data (a) and 
DSM based on the precise DBM (b) (bridge 2) 
Clearly, the bridge ROI DSM is refined in both cases; the 
surface and outlines of the bridges are smoother. For the test 
bridge 2, the “collapsed” bridge section due to tree crowns is 
well repaired. DSM with precise DBM can be used to improve 
the orthophoto generation of the highway corridor areas. In this 
paper, a new precise DBM generation method based on fusing 
LiDAR and aerial image data is introduced. The novel idea is to 
transfer smooth bridge boundaries extracted from the aerial 
image to the LiDAR data via the established co-registration 
between LiDAR intensity and aerial images. Note only 
approximate exterior orientation of the aerial image should be 
known. In case aerial imagery is not available, other high 
resolution satellite image can be also used. In this paper, the 
method is applied to produce precise DBM of straight and 
curved bridges by fusing LiDAR data and aerial images. In 
future research, complex multiple-layer bridge models using 
this methodology will be investigated. In addition, the 
possibility to apply this method to generate digital building 
model of complex building structures will be also explored. 
REFERENCE 
Comaniciu, D., Ramesh, V., Meer, P., 2003. Kernel-based 
object tracking. IEEE Transcations on Pattern Analysis and 
Machine Intelligence, Vol. 25, Issue 5, pp. 546-577 
Fischler, M. A., Bolles, R. C., 1981. Random sample consensus: 
a paradigm for model fitting with applications to image analysis 
and automated catography. Communications of ACM, Vol. 24, 
pp. 381-395 
Goepfert, J., Rottensteiner, F., 2010. Using building and bridge 
information for adapting roads to ALS data by means of 
network snakes. International Archives of Photogrammetry,