, Circle) 
  
riptor 
  
96 38 40 
8 length 
n 
  
, circle) 
E 
A 
M 
  
  
| 36 38 40 
3 length 
'ation 
  
  
es which 
re bigger. 
ter sepa- 
ed shape 
invariant 
a is used 
on of the 
the mo- 
ning the 
| process 
red com- 
xtraction 
o 0.2 sec 
sec (mo- 
t enough 
4. CONCLUSION 
The aim of the the investigation has been to deve- 
lop a simple but reliable procedure for the identi- 
fication and location of simple planar objects. Se- 
veral general questions were examined that arose 
during the recognition of simple and planar geo- 
metrical figures of an image sequence. Especially 
the structure of the object and the image motion 
recorded by an image sequence are exploited to 
solve the task. 
First a procedure was developed for the detec- 
tion and location of the border line in the image. 
For objects located in a natural environment the 
segmentation based on image motion was found 
to be the most promising procedure. The pro- 
cedure for extracting the region of interest uses 
the fact that a change can be identified in a dif- 
ference image. By that the process corresponds 
to classical procedures of the analysis of image 
sequences. 
A suitable procedure for reliable object recogni- 
tion has to be chosen from different existing pro- 
cedures. In our case of simple geometric fea- 
tures two procedures were suitable for finding 
affine-invariant quantities of a closed border line. 
These features are the form factor calculated from 
affine-invariant Fourier descriptors and the affine- 
invariant features determined by moments up to 
the third order. The sensitivity of the procedu- 
res with respect to noisy border lines is of special 
interest. The probability for misclassification ser- 
ves as a quality measure. Experiments with the 
three different types of objects are reported. For 
that the triangular, rectangular and circular bor- 
der lines of corresponding traffic signs are distor- 
ted and noise was added. The determination of 
the probability for a misclassification applies the 
maximum likelihood principle. The best results 
concerning the separability between the three ob- 
ject types are found by determining the affine- 
invariant features using the moments of an area 
which lies within a closed border line. Finally the 
identification of the object is done by maximum 
likelihood classification. For this purpose the four 
affine-invariant features can be used directly. 
REFERENCES 
Arbter, K. [1989]: Affine-invariant Fourier de- 
scriptors. In: From Pizels to Features, 
North-Holland, Amsterdam, 1989. 
Costa, M., Haralick, R., Phillips, T., Shapiro, L. 
[1989]: Optimal affine-invariant point mat- 
ching. In: Proc. of the DARPA Image 
Understanding Workshop, San Mateo, CA, 
1989. Morgan Kaufmann. 
Cyganski, D., Cott, T.A., Orr, J.A., Dodson, 
R.J. [1987]: Development, Implementation, 
Testing, and Application of an Affin Trans- 
form Invariant Curvature Function. In: Pro- 
ceedings of the 1st International Conference 
on Computer Vision, pp. 496-500, London, 
1987. IEEE Computer Society Press. 
Forsyth, D., Mundy, L., Zisserman, A., Brown, 
Ch. M. [1990]: Invariance - A New Frame- 
work for Vision. In: Proceedings of the 9rd 
International Conference on Computer Vi- 
sion, pp. 598-605, Osaka, 1990. IEEE Com- 
puter Society Press. 
Fukunaga, K. [1972]: Introduction to Statistical 
Pattern Recognition. Academic Press, New 
York, 1972. 
Geiselmann, Ch. [1992]: Zur automatischen 
Identifizierung einfacher Objekte aus einer 
Bildfolge. PhD thesis, Universität Stutt- 
gart, Institut für Photogrammetrie, Stutt- 
gart, 1992. | 
Hu, M.-K. [1962]: Visual Pattern Recognition 
by Moment Invariants. IRE Transactions on 
Information Theory, pp. 179-187, 1981. 
Jain, R. [1981]: Extraction of Motion Informa- 
tion from Peripheral Processes. IEEE Tran- 
sactions on Pattern Analysis and Machine 
Intelligence, vol. 3, no. 5, 1981. 
Naas, J., Schmid, H.L. [1972]: Mathemati- 
sches Wórterbuch. Akademie-Verlag, Berlin, 
Teubner Verlagsgesellschaft, Stuttgart, 1972. 
Novak, K. [1990]: Integration of a GPS-receiver 
and a stereo-vision system in a vehicle. In: 
International Archives of Photogrammetry 
and Remote Sensing, vol. 28, 5/1, pp. 16-23, 
Zürich, 1990. 
Schwarz, K.P. [1992]: Inertial Techniques in Ge- 
odesy - State of the Art and Trends. In: 
High Precision Navigation 91, Linkwitz, K., 
Hangleiter, U. (ed.), Dümmler Verlag, Bonn, 
1992. 
469