209
-jmhmhsb
Mann
In: Stilla U, Rottensteiner F, Paparoditis N (Eds) CMRT09. IAPRS, Vol. XXXVIII, Part 3/W4 — Paris, France, 3-4 September, 2009
6 RESULTS AND PERFORMANCE EVALUATION REFERENCES
The proposed algorithm is evaluated on a set of 1370 im
ages acquired in dense urban area with real traffic condi
tions. Figures 9-13 show some obtained results. In each
image the number of correct detection, false detection, and
true road signs are counted manually. We assume that if a
road sign is smaller than 10 pixels, we can not detect it.
We observed that there is 67% of good detection and 33%
of road signs are not detected. This is due to our camera
radiometric calibration problems that causes color detec
tion failure. As color detection is at the beginning of our
pipeline the shape detection and recognition processes are
not performed on the lost road signs.
The shape detection and recognition steps works well. We
mean that, in most of the cases they reject correctly the
false hypotheses and in the case of validation the type of
road signs are correctly distinguished. However, there is
5% of false detection. They are in most of the cases due
to the red lights behind the cars or the tricolor lights that
are very similar to wrong-way (see Figure 7(b)) traffic sign
(see Figure 9).
7 CONCLUSION AND TRENDS
In this paper we proposed a pipeline for road sign detection
in RGB image. Thanks to ellipse detection and rectifica
tion processes, the algorithm is not sensitive to road sign
orientation. The matching step provides a reliable recogni
tion of road sign type.
Evaluations revealed that, the detection rate is about 70%.
This is always due to failure in color detection step. Better
radiometric calibration of the camera and test of other color
spaces are the work in progress for improving color detec
tion. In contrast to color detection step our shape detection
and recognitions steps provide satisfactory and reliable re
sults.
The proposed algorithm can be easily extended to handle
the rectangular and triangular road signs. For this purpose,
it is enough to adapt the shape detection step and both other
steps remain unchanged.
In Figure 13 we can see a particular case which represent
two small road signs on a bigger road sign. These cases
can be handled using a stereo system allowing 3D position
and size estimation.
In real time applications such as driver assistance systems,
it is often interesting to track objects in video sequences.
Actually, our algorithm does not work in real time and can
not be applied on video sequences. The edge detection
is the most time consuming step. In order to reduce the
processing time, other edge detectors such as Sobel or Pre
witt filters can be applied and evaluated. The search area
can also be limited to remove the sky and so speed up the
global processing time.
Aly, F. and Alaa, A., 2004. Detection, categorization and
recognition of road signs for autonomous navigation. In:
Proceeding of Advanced Concepts for Intelligent Vision
System, Brussels, Belgium.
de la Escalera, A., Armingol, J., Pastor, J. and Rodriguez,
F., 2004. Visual sign information extraction and identifica
tion by deformable models for intelligent vehicles. IEEE
Transactions on Intelligent Transportation Systems 5(2),
pp. 57-68.
de la Escalera, A. Moreno, L. S. M. A. J., 1997. Road
traffic sign detection and classification. IEEE Transactions
on Industrial Electronics.
Deriche, R., 1987. Using canny’s criteria to derive a re
cursively implemented optimal edge detector. The Interna
tional Journal of Computer Vision 1(2), pp. 167-187.
Devemay, F., 1995. A non-maxima suppression method for
edge detection with sub-pixel accuracy. Technical Report
RR-2724, INRIA.
Fischler, M. A. and Bolles, R. C., 1981. Random sample
consensus: A paradigm for model fitting with applications
to image analysis and automated cartography. Communi
cations of the ACM 24(6), pp. 381-395.
Habib, A. and Jha, M„ 2007. Hypothesis generation
of instances of road signs in color imagery captured by
mobile mapping systems. International Archives of the
Photogrammetry, Remote Sensing and Spatial Information
Sciences 36 part 5/C55) pp. 159-165.
Ishizuka, Y. and Hirai, Y., 2004. Segmentation of road sign
symbols using opponent-color filters. In: ITSWC, Nagoya,
Japón.
Piccioli, G., Micheli, E. D., Parodi, P. and Campani, M.,
1996. Robust method for road sign detection and recogni
tion. Image Vision Comput. 14(3), pp. 209-223.
Priese, L., Lakmann, R. and Rehrmann, V., 1995.
Ideogramm identification in a realtime traffic sign recog
nition system. In: Proceeding of intelligent vehicles apos,
IEEE, Nagoya, Japón.
Reina, A. V., Sastre, R. J. L., Arroyo, S. L. and Jiménez,
P. G., 2006. Adaptive traffic road sign panels text extrac
tion. In: ISPRA’06: Proceedings of the 5th WSEAS Inter
national Conference on Signal Processing, Robotics and
Automation, World Scientific and Engineering Academy
and Society (WSEAS), Stevens Point, Wisconsin, USA,
pp. 295-300.
Rosin, P. L., 2003. Measuring shape: ellipticity, rectangu
larly, and triangularity. Machine Vision and Applications
14(3), pp. 172-184.
Shaposhnikov, D., Podladchikova, L., Golovan, A. and
Shevtsova, N., 2002. Road sign recognition by single,
positioning of space-variant sensor window. In: Proc.
15th International Conference on Vision Interface, Cal
gary, Canada, pp. 213-217.
