LE b 
Jg 
Xe 
Og 
World coordinate system 
   
  
GPS Antenna 
XK 
  
  
& 
| t 
LE > 
Xe 
Op 
World coordinate system 
  
  
  
(5 
Figure 4: Conceptual figures of Geo-referencing data 
sources (a) geo-referencing of range scan line, (b) geo- 
referencing of line image. 
2. Range point belonging to horizontal line segments and 
at ground elevation might be the measurement of road 
surface. Relative elevation from the origin of LD-A 
to the nearest ground surface is almost constant. It is 
calculated previously in calibration stage. 
3. Range point of tree always has a high distribution 
variance among the neighboring range points of suc- 
cessive range scan lines. 
4. Range point of small clusters always has low reliabil- 
ity. 
Geometric extraction of vertical building surfaces, ground 
surfaces, non-vertical building surfaces and trees is con- 
ducted in a subsequent way using the corresponding group 
of range points. On the other hand, classification of range 
points is validated and refined using the extracted geomet- 
ric features. For example, the range points near to a verti- 
cal planar face are classified to the measurement of verti- 
cal building surface. On the other hand, the range points 
belong to vertical line segments might be discarded, if no 
vertical planar face is extracted from the range points. In 
the extractions of vertical building surfaces, Z-image is 
exploited, which has been defined and demonstrated with 
efficiency in urban out-door area in Zhao and Shibasaki 
2000, so that a three-dimensional extraction is converted 
to a two-dimensional problem. Line segments are first ex- 
tracted from the Z-image of the range points belonging to 
the group of vertical building surface. Line segments are 
then recovered to vertical polygons using the correspond- 
ing range points to define boundaries (see Fig.6(b)). As 
for the extraction of ground surface, range points are pro- 
jected onto a regularly tessellated horizontal plane, and an 
elevation map of the ground surface is generated using the 
minimal Z-value of the range points in each grid cell. A 
TIN model is constructed on the elevation map to repre- 
sent the surface geometry (see Fig.6(a)). After extracting 
building and ground surface, the left set of range points 
are a mixed data of trees, parking cars, utility poles, ir- 
regular points and so forth. One of the major differences 
between trees and others is the measurement of trees al- 
ways yields a large cluster of range points. Z-image of the 
left set of range points is generated. The range points that 
correspond to isolated or small cluster of image features in 
Z-image are removed. Triangular cells are used to model 
trees, where each range point of the measurement of trees 
is amplify into a small triangle. Shape of the triangle is de- 
cided in a random way, whereas, surface normal is pointing 
to the center of the range scan line, and distances from each 
vertex to the range point are limited in a given range (see 
Fig.6(a)). 
S TEXTURE MAPPING 
Fig.7 explains the principle of line image measurement, 
where trajectory of the line camera #5, i.e. positions and 
principle axes, is shown in Fig.7(a), the corresponding strip 
of line images is shown in Fig.7(b). Distortion in the strip 
of line images is obvious. It is caused by the change of 
relative distance and direction from line camera to the ob- 
jects. In order to correct the distortions and generate a tex- 
ture of direct proportion to each object, line images are 
re-sampled as follows. 
1. Texture of each building surface is generated using 
the data of the line camera that both is close and has a 
large incident angle to the planar face. It is conducted 
as follows. Project and resample the line images of 
each line camera on the planar face separately, where 
the projection vector of each line image pixel is cal- 
culated using the formula 3. The texture that is made 
of the maximal line image pixels is selected. 
2. Texture of TIN-based ground surface is generated in 
two steps. First projecting and re-sampling the line 
images of all line cameras on a horizontal plane at al- 
most the same elevation level with the ground surface, 
where image pixels of line camera #2 and #5 have 
higher priorities and will not be covered by those of 
other line cameras. Secondly projecting each vertex 
of TIN model along the direction of range beams to 
the horizontal plane to create a connection between 
TIN-model and the texture image. 
3. Texture of trees is generated using synthetic colors, 
since the projection of line images onto the range points 
of trees is not reliable due to occlusions. 
An example of texture mapping onto geometric model of 
Fig.6(b) is shown in Fig.7(c). 
—122- 
6 E 
An ex 
ronme 
measu 
of 20k 
hicle t 
lines : 
and li 
All ra 
using 
tured 
recon: 
using 
ment, 
meası 
surfac 
ping i 
line ii 
mode 
the gi 
are cc 
sor, C 
proce 
7 C 
In th 
gener 
ment 
eras. 
chror 
tem, 
feren 
syste 
door 
range 
of ur 
samp 
it is « 
recor 
ACK 
The : 
Air “