Table 1. Specification of omnidirectional Camera 
  
  
  
  
  
  
  
  
Focal Length 5.9 [mm] 
Image Circle ¢ 18[mm] 
Image Size 3032x2008 [pix] 
Shutter Interval 2 [sec/frame] 
Color Depth 24/32[bit/pix] 
  
  
Figure 3. Example of omnidirectional image 
2.3 Real-time positioning by D-GPS and GPS interpolation 
device 
When using GPS equipment in an urban area, the quality of 
positioning can drop due to shielding and multi-path effects 
caused by buildings and other objects. In recognition of such 
problems, we have adopted a D-GPS device having a multi- 
path-elimination function. In addition, to deal with shielding 
effects that not even GPS equipment with the above function 
can eliminate, we have also developed for our system a GPS 
interpolation device that employs a gyro sensor and speed 
sensor. A key feature of this device is its ability to perform 
positioning by interpolating in real time while directly inputting 
data from GPS equipment and monitoring the quality of GPS 
data. Figure 4 shows an example of actual position data for a 
certain area. For comparison purposes, the figure also shows 
typical position data from GPS equipment. Specifically, the 
solid circles indicate typical GPS position data with no 
interpolation while the empty circles indicate position data 
obtained by our system. These circles draw a curve initially 
moving downward from the left and then changing direction 
and moving upward to the right. Examining this data, we see 
that typical GPS position data is missing in the "lacking 
section" while position data by our system features interpolated 
data in the “interpolated section." In addition, typical GPS 
position data includes disorder as a result of poor quality data in 
the “noisy area." The results demonstrate that our mobile 
mapping system can obtain  high-accuracy positional 
information in a stable manner even under conditions that 
prevent positioning by GPS. 
  
Fu Gu X S 
—Q— 
       
Typical D-GPS system |; 
—o— Our system | 
DS iu 
     
  
H^ 
  
ue, x, ^ 
eere gh 
  
  
  
Figure 4. Comparison our system's position data with typical D- 
GPS data 
2.4 Synchronization of positional information and 
omnidirectional images 
If mistakes are made when assigning correspondence between 
omnidirectional images and positional information, it will not 
be possible to correctly convert reconstructed building models 
to global coordinates. It is therefore important to establish strict 
synchronization between GPS-based positional information and 
captured omnidirectional images. In the former system that we 
created, the lag between shooting time and positioning time was 
a maximum of 8 seconds due to Windows Operating System 
delays as a result of image-storage processing. In our current 
system, however, the positioning computer records the shutter 
opening time of an omnidirectional camera in GPS time. As a 
result, the lag between shooting time and positioning time has 
been reduced to within 0.5 second in our system. 
3. BUILDING-MODEL RECONSTRUCTION BY 
OMNIDIRECTIONAL STEREO VISION 
Figure 5 shows the process flow from image input to building- 
model reconstruction. In this process, the system first 
determines camera attitude with respect to a captured image by 
estimating the focus of expansion (FOE) position within the 
omnidirectional image. Then, with respect to the images 
corrected for camera attitude, the system extracts and tracks 
feature points targeted for measurement and calculates their 
three-dimensional positions by stereo measurements. Next, the 
system configures surfaces using the calculated three- 
dimensional points and reconstructs building shape. Finally, the 
system re-projects the reconstructed shape onto the building's 
lateral omnidirectional image and extracts texture to reconstruct 
the building model. 
The following describes the above process in more detail. 
—200— 
  
3.1 
syst 
Tc 
the 
pos 
bef 
pos 
the 
mai 
tha: 
mil 
pos 
ins! 
of 
nez 
it f 
has 
im: