den ten nee 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part BS. Istanbul 2004 
in elevation. These results seems somehow too good, so we 
want to review the likelihood of the image point distribution in 
the simulation, especially as far as the number of ray per image 
is concerned. Anyway, if the trend will be roughly confirmed, 
some conclusion can be drawn: using GPS code solution should 
not help much, since the accuracy of this solution, under the 
extreme condition (4 satellites, poor signal reception) we are 
considering, cannot be expected to be better than 10 cm. On the 
other hand, 50 cm is well within the tolerance of points taken 
by 1:2000 maps, but too small for the average accuracy of 
1:5.000 maps for the planimetry. Since 1:2.000 maps are 
restricted to urban areas, also this method may not lead to 
significant improvements of the length of the recoverable loss 
of lock. 
5. CONCLUSIONS AND FUTURE WORK 
Though we cannot claim a verified accuracy level, we are 
confident that the proposed methodology, as shown in the test 
images, is capable to achieve reliable result, at least for short 
sequences. Since to ensure a good motion estimation good 
interest points well distributed in the scene are necessary, the 
method may work well especially in sub-urban environments, 
where buildings and man made objects make the tracking 
process easier and the loss of lock should be short in time. 
There are still a few steps of the process where there significant 
improvements are possible and necessary. The most important 
is the evaluation of the putative correspondences: based just on 
a disparity threshold, is quite rudimental and spends a lot of 
computation power providing results of different quality 
depending on the scene characteristics. Since an important 
problem is still to find correspondences in the bottom part of the 
images, the use of a LSM algorithm may possibly allow to 
incorporate more points at the image bottom, where scale 
changes more dramatically between pair of frames. Introducing 
a further guided matching, once the camera pose is correctly 
estimated, in order to gain more correspondent points, it's not 
unreasonable. Since finding points on the paved road surface is 
not too easy due to the difficulty to adjust to changing lightning 
conditions and shutter speed, another possible improvement is 
to apply some sharpening filter such as the Wallis filter. 
Finally, the management of the various blocks may be 
optimized considering different schemes and made more 
flexible. 
As already mentioned, the first step will be to come up with 
figures on the accuracy of the reconstruction in object space. 
This may also put some light on the possibility to use the 
procedure for improvement of the exterior orientation computed 
by GPS only. To this aim, the GPS solution will provide initial 
values to speed up the search for correspondences, while the 
trifocal geometry will serve as filtering of outliers. Finally, an 
integrated bundle block adjustment including the GPS solution 
as pseudo-observed values may provide an improved block 
geometry. 
6. REFERENCES 
References from Journals: 
Crosilla, F., Visintini, D., 1998. External Orientation of a 
Mobile Sensor System for Cartographic updating by dynamic 
vision of digital map points. Bollettino di Geodesia e Scienze 
Affini, 1 (1998), pp. 41-68. 
Fischler M., Bolles R., 1981. Random sample consensus: a 
paradigm for model fitting with application to image analysis 
and automated cartography. Commun. Assoc. Comp. Mach., 
vol. 24:3, pp. 81-95. 
Hom, B.KP. 1987. Closed-form solution of absolute 
orientation using unit quaternions, J. Opt. Soc. Amer. A vol. 4, 
no. 4, pp. 629-642. 
Longuet Higgins, H. C., 1981 A computer algorithm for recon- 
structing a scene from two projections. Nature, pp. 133-135, 
September 
Quan, L., Lan, Z., 1999 Linear N-Point Camera Pose 
Determination in /EEE Transactions on Pattern Analysis and 
Machine Intelligence vol. 21, No. 8, August 
Torr, P.H.S., Zisserman, A. 2000. MLESAC: A new robust 
estimator with application to estimating image geometry. In 
Computer Vision and Image Understanding, 78(1), pp. 138- 
156. 
References from Books: 
Hartley, R., Zisserman, A., 2000. Multiple View Geometry in 
computer vision. Cambridge University Press, Cambridge, pp. 
1-496. 
Cooper M.A.R., Robson, S., in Atkinson, K.B., 1996. Close 
Range Photogrammetry and Machine Vision. Whittles 
Publishing, Caithness, pp. 9-50. 
References from Other Literature: 
Brown, M., Lowe, D.G., 2002. Invariant Features from Interest 
Point Groups British Machine Vision Conference, BMVC, 
Cardiff, Wales. 
Forlani G., Pinto L., 1994. Experiences of combined block 
adjustment with GPS data. Int. Archives of Photogrammetry 
and Remote Sensing, Vol. 30 Part 3/1, Muenchen, 1994, 219- 
226. 
Fórstner, W. and E. Gülch, 1987. A fast operator for detection 
and precise location of distinct points, corners and centres of 
circular features. In: ZSPRS Intercommission Workshop, 
Interlaken, pp. 149-155. 
Harris C., Stephens M. 1987. A combined corner and edge 
detector. In: Proceedings of the Alvey Conference, pp. 189- 
192. 
Hartley, R., Sturm, P., 1994. Triangulation. In Proc. DARPA 
Image Understanding Workshop, pp. 745-753. 
Heipke C. et al, 2002. The OEEPE test Integrated Sensor 
Orientation, OEEPE official publication n. 43. 
Hu, M., Yuan, B. 2002 Robust Estimation of Trifocal Tensor 
using Messy Genetic Algorithm. In Pattern Recognition. 2002. 
In: Proceedings of the 16th International Conference on Pattern 
Recognition. 
Pollefeys, M., 1999. Self-calibration and metric 3D 
reconstruction from uncalibrated image sequences. PhD thesis. 
EAST PSI, K.U. Leuven. 
Tao, C., Chapman, M. A., El-Sheimy, N. and Chaplin, B.. 1999. 
Towards automated processing of mobile mapping image 
sequences. International Archives of Photogrammelry and 
Remote Sensing, Vol. XXXII, 2W 1 
eo