156 
Although at the time only pseudorange measurements 
are collected, to improve the accuracy of GPS positio 
ning these data are differentially corrected in post 
processing together with the data recorded by a base 
station receiver of same type as the one on the van. The 
position of any object, visible on a pair of simultaneous 
images, can be determined firstly in the reference sys 
tem of the two cameras, using digital photogrammetric 
techniques and computer vision algorithms, and after, 
merging this information with the GPS corrected data, 
it can be reported on any user selectable map. 
As the main goal of our project was to realize a low 
cost mobile mapping system, GeoVision is not equip 
ped with an Inertial Navigation System (INS). There 
fore, as final step of the whole georeferencing process, 
our system doesn't provide 3D coordinates of object 
position, but rather its 2D coordinates on a map, as 
depicted in Fig. 3. 
Given the absence of the INS, our major efforts were 
spent to implement a mobile system capable to carry 
out, anyway, rapid and enough accurate road surveys 
for all those public or private agencies that are intere 
sted in road data collec-tion for GIS applications. We 
have also in plan to employ such a system for teaching 
purposes in the Topographic field, in order to put in 
touch our students with GPS technology, digital photo- 
grammetry and related digital image processing techni 
ques and GIS applications. 
Fig. 2: External view of the van 
From software point of view, the GeoVision system 
was developed as four different, but not separated, 
modules, namely: Caltool, for the calibration of each 
CCD camera [6], StereoCalib, for the calibration of the 
whole stereoscopic system, EpSearch for the image 
matching and the triangulation step, i.e. the determina 
tion of 3D object coordinates in the stereoscopic refe 
rence system [2], and finally NavPos, the software 
module which integrates the corrected GPS data with 
the results of triangulation in order to determine the 
object position on a 2D map (Fig. 3). 
In this paper we will focus the reader's attention mainly 
on this last module, pointing out the structure of the 
algorithm implemented in order to recover the van 
orientation on a mapping plane using only the GPS 
data and a Kalman filter. 
Fig. 3: The whole georeferencing process 
2. THE GEOREFERENCING PROCESS 
As exposed in [6], assuming that the origin of the 
stereopair is located on optical center of the left CCD 
camera, respect to the moving direction (Fig. 4), the 3D 
position of an object in the scene can be calculated as 
follows: 
1) Compute the 3D object position in rover frame (X r ) 
by rototraslation from camera frame (Z c ), as 
depicted in Fig. 5a and 5b, 
r/'(o = r; + R:- r ;(o (i) 
2) Compute the object position in the mapping frame 
(X m ) by rototraslation from rover frame (E r ), as 
depicted in Fig. 6, 
r m (0 = r;(o + R;"(o-r/(o ( 2)