ISPRS Commission II, Vol.34, Part 3A „Photogrammetric Computer Vision“, Graz, 2002 
  
are at the level of 2-3 cm for position coordinates, and ^10 
arcsec and 10-20 arcsec for attitude and heading components, 
respectively. These naturally do not represent the final 
mapping accuracy, which can only be confirmed by an 
independent comparison with the ground control, as 
presented in Section 6. Figure 1 illustrates the system 
architecture (Cairo or Budapest Grejner-Brzezinska and Toth, 
2002), and Figure 2 presents the prototype hardware 
configuration. 
ODOT Centerline Surveying System 
Hardware Configuration 
GPS 
Antenna 
Trimble 
4000ssi 
   
       
   
  
    
  
  
Pulnix 
TMC-6700 
VGA, Mouse N 
Figure 1. Design architecture and data processing flow. 
  
  
  
Camera CCD pixel size 9 micron 
Camera focal length 6.5 mm 
Camera height above road surface 3m 
  
Image scale 3/0.0065=461 
  
  
  
Ground pixel size at nadir (no tilt) 4.1 mm 
Ground coverage along vehicle 2.68 m 
Ground coverage across vehicle 2m 
  
Max speed, no overlap at 10 FPS 26.8 m/s (96 km/h) 
13.4 m/s (48 km/h) 
  
Max speed at 50% overlap 
  
  
  
  
Table 1. Sensor characteristics and the image acquisition 
parameters. 
The imaging module consists of a single, down-looking, 
color digital camera, Pulnix TMC-6700, based on 644 by 482 
CCD, with an image acquisition rate of up to 30 Hz (10 Hz is 
the target for our application), which allows for full image 
coverage at normal highway speed or 5096 image overlap at 
reduced speed (footprint size is about 2.68 by 2 m; see Table 
1). More details are provided in (Grejner-Brzezinska and 
Toth, 2000; Toth and Grejner-Brzezinska, 2001 a and b). The 
imaging system provides a direct connection between the 
vehicle georeferencing (positioning) module and the road 
marks visible in the imagery, allowing for the transfer of the 
coordinates from the reference point of the positioning 
system (center of the INS body frame) to the ground features. 
Naturally, calibration components, including the camera 
interior orientation (IO), as well as INS/camera boresight 
calibration components are needed (for algorithmic details 
see, for example, Grejner-Brzezinska, 2001). For 3D image 
processing, a 50-6096 overlap is needed along the vehicle 
motion, which can be achieved with the hardware 
implemented in our system. Stereovision is realized by the 
platform motion, which, in turn, emphasizes the need for 
A - 363 
high-precision sensor orientation provided by direct 
georeferencing. Table 1 summarizes the camera 
characteristics and the image acquisition conditions. 
ODOT District 1 office built the complete system with all the 
sensors and supporting hardware installed in early 2002 and 
Figure 2 shows the surveying vehicle. 
  
Figure 2. Mapping vehicle. 
3. IMAGE SEQUENCE PROCESSING CONCEPT 
There are two key questions regarding the development of 
the image sequence-processing concept. The first is whether 
a more complex stereo model-based technique or a simple 
monoscopic method should be used for the centerline 
position extraction process. Second is the question of 
whether a completely real-time (or near real-time) solution 
implementation should be considered or whether post- 
processing should remain the only option. The main goal of 
on-the-fly image processing is to determine the centerline 
image coordinates in real time, so that only the extracted 
polyline, representing the center/edge lines, would be stored 
without a need to store the entire image sequence. Clearly, 
there is a strong dependency among these options and the 
decision was made at the beginning to develop the system 
with full 3D capabilities in a possibly real-time 
implementation. Later, based on the initial performance, the 
design may be modified. In simple terms, the stereo 
processing can provide excellent accuracy but it imposes 
more restrictions on the data acquisition process, such as the 
need for continuous image coverage with sufficient overlap, 
and it definitely requires more resources. The single image 
solution however, is a compromise in terms of accuracy but it 
is very tolerant toward the image acquisition process, e.g., 
gaps will not cause any problems and the processing 
requirements are rather moderate. 
The real-time image processing is technically feasible due to 
the simple sensor geometry and the limited complexity of the 
imagery collected (single down-looking camera acquiring 
consecutive images with about 50% overlap; only linear 
features are of interest). The most challenging task is the 
extraction of some feature points around the centerline area, 
which can be then subsequently used for image matching. 
Note that the availability of the relative orientation between 
the two consecutive images considerably decreases the search 
time for conjugate entities in the image pair, since the usually 
2D search space can be theoretically reduced to one 
dimension, along the epipolar lines. However, errors in