obtained in different viewpoints are registered and inte- 
grated, and a completed model of urban environment is 
reconstructed. There are several drawbacks of stationary 
systems. First, in data acquisition, successive range views 
have to keep a degree of overlay, so that location and di- 
rection of viewpoints can be traced (or refined) by regis- 
tering range data. Planning for viewpoints and directions 
in data acquisition becomes difficult when measuring large 
and complicated scene, since a balance between the de- 
gree of overlay and the number of viewpoints has to be 
decided according to both target objects and registration 
method. Secondly, there is still no registration method that 
could succeed in automatically registering range data of 
all kinds. When the number of range views increases, reg- 
istration while keeping necessary accuracy becomes diffi- 
cult. Updating stationary systems to moving platform ones 
(called vehicle-borne system) for reconstructing 3D mod- 
els of large real scenes are very important. 
2 OUTLINE OF THE RESEARCH 
In the sensor system developed by Konno et al. 2000, three 
single-row laser range scanners and six line cameras are 
mounted on a measure vehicle (GeoMaster), which has 
been equipped with a GPS/INS/Odometer based naviga- 
tion system. The sensor system outputs three kinds of data 
sources. They are laser range data, line images, and nav- 
igation data. Either laser range data or line images are in 
the sensor's local coordinate system at the moment of mea- 
surement. They are synchronized with the navigation data 
using the sensors' local clock. This research contributes 
to a method of reconstructing textured 3D model of ur- 
ban out-door environment by fusing the data outputs of the 
sensor system. It has two procedures. A geometrical sur- 
face model is first generated using the integrated model of 
laser range data, where laser range data in the sensor's lo- 
cal coordinate system at the moment of measurement are 
geo-referenced to a world coordinate system using both the 
navigation data and the calibration parameters of the sen- 
sor system. Texture data are then generated by projecting 
line images onto the surface model, where line images are 
geo-referenced to a world coordinate system in the same 
way with that of laser range data. In the following, we will 
first briefly describe the hardware system and the way for 
geo-referencing each kind of data sources. We then present 
the method for surface model reconstruction and texture 
mapping. An out-door experiment is conducted in a real 
urban environment, Kirigaoka Apartment Complex, where 
the measurement vehicle ran a course about 1.5km at a 
speed of 20~40km/h. A textured CAD model including 
the urban features like buildings, trees, roads etc. along the 
measurement route is generated in a full-automated mode. 
The experiment results and discussions are presented sub- 
sequently. 
3 SENSORSYSTEM AND GEO-REFERENCING OF 
DATA OUTPUTS 
The sensor system consists of three different kinds of sen- 
sors and each for a specific purpose. They are laser range 
Scanning 
— 
by 
À 
== 
   
i 
; | d i 
Ground surface 
  
n 
Building Surface ES = 
scanning sequence 
> nn (c) 
i E Te Seer Position Figur 
yer ey lind Area 
A 
(b) Ground Surface 
* 
Figure 1: Laser range finder (LD-A) and examples of range 
data (a) configuration of LD-A, (b) range points in a scan 
ol 
line, (c) a piece of range image. 
scanners - the sensor for measuring object geometry, line 
cameras - the sensor for capturing object texture, and Geo- 
Master - the moving platform. 
LD-A£ 
3.1 Laser range scanner - Sensor for measuring ob- f 
ject geometry 
Single-row laser range scanners, LD-A, produced by IBEO 
Lasertechnik, are exploited in the sensor system (see Fig.1(a)). Figur 
In one scanning (a range scan line), LD-A profiles 480 ment 
range points of the surroundings on the scanning plane 
within 300 degrees. A blind area of 60 degree exists due urbar 
to the hardware configuration (see Fig.1(b)). LD-A has a this 
maximum range distance of 100 meter and an average er- platf 
ror of 3cm. Frequency of LD-A is 20Hz, implying that it 34 
profiles 20 range scan lines per second. Fig.1(c) shows a ; 
piece of range image, where each column corresponds to Thre 
a range scan line, and range scan lines are aligned in the of Gi 
order of measurement sequence. 
came 
ent 
3.2 Line camera - sensor for capturing object texture rio a 
; ; betw 
Line cameras are implemented in the sensor system. Each fhrou 
has a 8mm F4 fish-eye lense with a vision field of 180 de- same 
gree on it (see Fig.2(a)). In each snapshot, a single-row Sens 
image (line image) of 1 x 2048 pixels is captured on the SOS. 
scanning plane. Among the 2048 pixels, about 224 pixels Whe 
(= 20 degree) on each side are discarded due to high lens ages 
distortion. Line images are captured at a rate of 80Hz by line. 
each line camera. Fig.2 shows a strip of the line images, abou 
where each column corresponds to the valid pixels of a line and. 
image. of H 
data 
3.3 GeoMaster - moving platform throi 
The measurement vehicle (Fig.3(b)) - GeoMaster is equipped 3.5 
with a high accurate GPS/INS/Odometer based navigation 
system - HISS (Konno et al. 2000). A method of improv- Fig. 
scan 
ing the accuracy of the navigation system in highly dense 
—120-