i 1 
| surface 
— 
y sequence 
© 
nples of range 
ints in a scan 
geometry, line 
ture, and Geo- 
1easuring ob- 
luced by IBEO 
em (see Fig.1(a)). 
\ profiles 480 
canning plane 
'ree exists due 
). LD-A has a 
an average er- 
mplying that it 
g.1(c) shows a 
corresponds to 
aligned in the 
object texture 
r system. Each 
ield of 180 de- 
t, a single-row 
aptured on the 
yout 224 pixels 
ue to high lens 
ate of 80Hz by 
he line images, 
pixels of a line 
[aster is equipped 
ased navigation 
hod of improv- 
in highly dense 
  
Fish-eye lens 
(8mm F4, 180?) 
       
i ee oe 
E 
qu, 
5 
Line camera 
NE 
ene 
Line image sequence 
í 
A 
Figure 2: Line Camera and an example of line image strip 
GPS Antenna Line Cameras 
d 
  
  
QINS/GPS ——— 
Vehicle direction 
C SA Line Camera 
LD-Ag1& S HS kL 
LD-AHIN #2 #0 
   
V 
(a) (b) 
Figure 3: Sensor system (a) sensor alignment, (b) measure- 
ment vehicle (GeoMaster). 
urban area is addressed in Zhao and Shibasaki 2001. In 
this paper, we assume the navigation error of the moving 
platform is small enough to be neglected. 
3.4 Sensor alignment 
Three LD-As and six line cameras are mounted on the roof 
of GeoMaster as shown in Fig.3(a). Both LD-As and line 
cameras are installed with their scanning planes at differ- 
ent angles to reduce occlusion. In this research, all exte- 
rior calibration parameters (relative angles and distances) 
between the sensors' local coordinate system are obtained 
through physical measurement; all interior calibration pa- 
rameters (e.g. focus length) are obtained from maker or 
sensors' specification. For data measurement, all the sen- 
sors keep recording data sources as the vehicle moves ahead. 
When GeoMaster moves at a speed of 20km/h, line im- 
ages are captured at an interval of about 6.9cm by each 
line camera, range scan lines are profiled at an interval of 
about 27.8cm by each LD-A, and navigation data (location 
and direction of the vehicle at the local coordinate system 
of HISS) are measured at an interval of 20cm. Navigation 
data is associated to each line image and range scan line 
through linear interpolation using the sensors' local clock. 
3.5 Geo-referencing data sources 
Fig.4 shows the conceptual figures of geo-referencing range 
scan lines and line images in each sensor’s local coordinate 
   
  
systems at the moment of measurement to a world coordi- 
nate system. According to the navigation data associated to 
each range scan line and line image, a transformation ma- 
trix T5, from the coordinate system of HISS to a world co- 
ordinate system is calculated, where the origin of HISS is 
at the center of GPS antenna. On the other hand, a transfor- 
mation matrix 7}; from the coordinate system of LD-A and 
a transformation matrix 7;;, from the coordinate system of 
line camera to the coordinate system of HISS are calcu- 
lated based on the exterior calibration parameters. Both 
Thg» Tin and Ten are in homogenous notations. In the case 
of LD-A, a right-hand coordinate system is defined with 
its origin at the center of the laser head. The laser head 
of LD-A does a clockwise rotation with its starting angle 
downward. Z-axis and X-axis are defined composing the 
scanning plane, at the angles of 180° and 270° from the 
starting angle. A range point with a range distance of r at 
the angle of « is geo-referenced to the world coor-dinate 
system as follow, 
(r,9, 2, 1)7.— Tnglin(—r sina, 0, —r cosa, DT. uw) 
In the case of line camera, a right-hand coordinate system 
is defined with its origin at its projection center. Z-axis is 
defined to the upward of the line camera, and Y-axis is re- 
verse to the projection axis. Focus length (f) of the line 
camera as well as a formula defining the relationship be- 
tween the index of image pixel (h) and its projection angle 
(w) towards the projection axis is obtained from the sen- 
sor’s specification. 
w = 2 x arcsin((h — o)/2/ f). (2) 
where o is the image center, which is obtained by doing 
physical measurement using sample images. Using for- 
mula 2, the projection vector of image pixel (A) is geo- 
referenced to the world coordinate system as follows, 
7.9.2,0 T =T,,T-1(0, — cosw,sinw, 0 T. (3) 
g 
4 CREATING GEOMETRIC MODEL 
The goal of geometric modeling is to generate a surface 
representation of urban features with small data size for the 
use of commercial applications. In the present research, 
urban features are reconstructed in three levels. They are 
buildings, trees, and ground surfaces, where buildings are 
further divided into vertical and non-vertical building sur- 
faces. Geometric modeling is conducted in two interrelated 
procedures. They are classification of range points and ge- 
ometric feature extraction. Range points are classified by 
first segmenting each range scan lines into line segments, 
then grouping the range points in a hierarchical procedure 
based on the following simple and intuitive rules. 
1. Range point belonging to vertical line segments of 
range scan line might be the measurement of vertical 
building surface. 
—121-