Valentine Vega
, D
e Sidib
and Yohan Fougerolle
Universitas Gunadarma, Jalan Margonda Raya 100, Depok, 16424 West Java, Indonesia
e de Bourgogne, Laboratoire Le2i, UMR CNRS 5158, 12 Rue de la Fonderie, 71200 Le Creusot, France
Road Sign Detection, Color Segmentation, Contour Fitting, Gielis Curves.
Road signs are among the most important navigation tools in transportation systems. The identification of
road signs in images is usually based on first detecting road signs location using color and shape information.
In this paper, we introduce such a two-stage detection method. Road signs are located in images based on
color segmentation, and their corresponding shape is retrieved using a unified shape representation based on
Gielis curves. The contribution of our approach is the shape reconstruction method which permits to detect
any common road sign shape, i.e. circle, triangle, rectangle and octagon, by a single algorithm without any
training phase. Experimental results with a dataset of 130 images containing 174 road signs of various shapes,
show an accurate detection and a correct shape retrieval rate of 81.01% and 80.85% respectively.
Intelligent Transportation System (ITS) has been a
rapidly growing topic of research over the last years.
One important component of any ITS is the capability
of automatic road sign recognition. The detection and
identification of road sign is useful in highway main-
tenance, sign inventory or driver support systems.
A road sign recognition system is composed of
two main stages: the detection and the recogni-
tion (Bascon et al., 2007). Detection is performed
to obtain an initial candidate of road sign, i.e. the
possible region that represents road sign characteris-
tics, using either color or shape information. Conse-
quently, it requires a recognizer equipped with a set
of features or patterns (Bascon et al., 2007; Escalera
et al., 2003). Commonly used classifiers include neu-
ral network (Bascon et al., 2007), clustering (Escalera
et al., 2003), nearest neighbor (Escalera et al., 2004),
Laplace kernel (Fang et al., 2003), and support vector
machines (Nguwi and Kouzani, 2008).
In this paper we focus on the detection stage of
road signs in still images or videos. Several prob-
lems on detecting road signs in natural scene images
include misdetection, occlusions, motions blur issues
and scale variations.
The two main features used to detect and recog-
nize road signs are the color and the shape. Color is a
powerful cue for object detection, but is sensitive to
image acquisition conditions. On the other hand, the
shape is invariant to illumination, but sensitive to oc-
clusions and perspective distortions. Color based de-
tection methods are based on color segmentation. For
example, (Varun et al., 2007) use a simple threshold
formula applied to the red color channel to detect red
road signs. Several color spaces can be used, includ-
ing HSV (Paclik et al., 2000), HSI (Escalera et al.,
2003; Fang et al., 2003), CIECAM97 (Gao et al.,
2006) and IHSI (Fleyeh, 2004).
Shape based methods have strong robustness to
changing illumination as they detect shapes based on
edges. The most recurrent method is certainly the
generalized Hough transform (Ballard, 1981). Re-
cently, (Loy and Zelinsky, 2002) propose the fast ra-
dial symmetry transform which is a technique similar
to Hough transform. The method was extended and
successfully used to detect regular polygons by (Loy
and Barnes, 2004). (Gavrila, 1999) uses a template
matching approach for shape detection. First, edges in
the original image are found. Then, a distance trans-
form image is created and match against a reference
template (for instance, a triangle).
In this paper, we propose a robust road sign detec-
tion method which uses the color information to local-
ize potential road signs and then, based on shape rep-
resentation using Gielis curves (Gielis, 2003), iden-
tifies the shape of the detected signs. The major
advantage of our approach is that it does not need
Véga V., Sidibé D. and Fougerolle Y..
DOI: 10.5220/0003802003930396
In Proceedings of the International Conference on Computer Vision Theory and Applications (VISAPP-2012), pages 393-396
ISBN: 978-989-8565-03-7
2012 SCITEPRESS (Science and Technology Publications, Lda.)
a multi-layer architecture as used in machine learn-
ing approaches (neural network or support vector ma-
chines). Furthermore, no initial training is needed.
Hence, the method is simple, fast, and is able to iden-
tify any common road sign shape (triangle, rectan-
gle, octagon and circle). Additionally, the proposed
method is scale invariant and accurately detects road
signs of different sizes.
The paper is organized as follows: The color seg-
mentation method adopted in our work is then de-
scribed in Section 2. Section 2 introduces Gielis
curves and the road sign shape identification algo-
rithm. Experimental results are shown in Section 3
before our conclusions in the last section.
The first step of the proposed method is the local-
ization of potential road signs in the image through
color segmentation. For robustness against lighting
variations, the Improved Hue, Luminance and Satu-
ration (IHLS) color space is selected. Once an image
is converted to IHLS color space, potential road signs
are detected using the segmentation method first in-
troduced by (Escalera et al., 2003) and also used by
(Fleyeh, 2004).
Figure 1 shows some segmentation results. As
can be seen, road signs are correctly localized with
the segmentation method in most cases. However, in
some situations, because of lighting changes or occlu-
sion, road signs are not entirely detected. Thus, some
post-processing steps are necessary to help the shape
reconstruction algorithm.
Introduced by (Gielis, 2003), the superformula ex-
tends the superellipses by introducing variable rota-
tional symmetry and asymmetric shape coefficients.
The angle φ is replaced by
to obtain m rotational
symmetries, and the unique shape coefficient n for su-
perellipses is replaced by a triplet (n
, n
, n
), leading
to the following radial polar parameterization:
r(φ) =
. (1)
a, b, and n
are positive real numbers (a, b, n
a and b control the scale while n
, n
, and n
trol the shape. m R
represents the number of ro-
tational symmetries. By modifying the shape param-
eters, the set of the regular polygons can be gener-
ated. Three additional coefficients T
, T
and φ
, cor-
responding to the position of the shape in the image
Figure 1: Example of segmentation results.
and its angular offset are added to represent Gielis
curves in the general case. Consequently, the pa-
rameters set Λ of a Gielis curve is defined as Λ =
{a, b, n
, n
, n
, m, T
, T
, φ
The process of 2D data representation by
Gielis curves starts by building a signed potential
field (Fougerolle et al., 2005). A potential field is a
signed function F which characterizes the inside and
outside of a closed object, such that F(x) is positive
if a point x lies within the object, F(x) is negative if x
lies on the outside, and F(x) = 0 if x lies on the curve.
The potential field is used to build a cost func-
tion to be minimized in the least-squared sense using
Levenberg-Marquardt (LM) algorithm as illustrated
in Figure 2.
Figure 2: Levenberg-Marquardt: evolution of the curve.
We evaluate the performance of the proposed ap-
proach on a dataset of 130 images containing 174
signs of various shapes and types. The dataset is a
subset of the one used by (Bascon et al., 2007).
Figures 3, 4, 5, and 6 show some detection re-
sults for circular, octogonal, rectangular and triangu-
VISAPP 2012 - International Conference on Computer Vision Theory and Applications
lar signs, respectively. As can be seen in these im-
ages, the shape reconstruction method can success-
fully identify different road sign types. It is important
to note that the algorithm accurately detects road sign
of different sizes in the image. This is a clear advan-
tage over the method developped by (Loy and Barnes,
2004) where different radii are tried in order to find
the correct size of road signs.
Figure 3: Detection for circular road signs.
Figure 4: Detection for octogonal road signs.
Figure 5: Detection for rectangular road signs.
Detection results using the entire dataset are sum-
marized in Table 1. From the 174 road signs, 141 are
correctly detected and 114 road sign shapes are suc-
cessfully reconstructed. Undesired detection occurs
for 23 objects having similar color with road signs,
while 27 road signs are incorrectly reconstructed. The
percentage of road sign detection and shape recon-
struction results are listed in Table 1. For each refer-
ence color, we calculate the percentage from the num-
ber of detected (reconstructed) road signs divided by
the total number of road signs of that particular color.
The total percentage is calculated from the number of
detected (reconstructed) road signs in both colors di-
vided by the total number of road signs in the images.
Note that the performance of the shape reconstruc-
tion method (identification of road sign shape) is di-
rectly related to the performance of the detection step
(color segmentation). Indeed, the shape of a road sign
which is not detected in the segmentation step will ob-
viously not be reconstructed. Several road signs in the
dataset appear at a far distance from the camera, and
can be eliminated in the post-processing stage, lower-
ing detection rate.
Incorrect reconstructions are due to different
causes including incorrect segmentation, wrong ini-
tialization of the fitting algorithm and perspective dis-
tortion as illustrated in Figure 7.
Figure 6: Detection for triangular road signs.
(a) Multiple parts extraction.
(b) High perspective distorsions.
(c) Intense light variations.
(d) Occlusions.
Figure 7: Examples of wrong reconstruction.
Table 1: Summary of detection and reconstruction results.
Detection (%) Reconstruction (%)
detected not detected correct incorrect
Red Signs 81.43 18.57 81.58 18.42
Blue Signs 79.41 20.59 77.78 22.22
TOTAL 81.03 18.97 80.85 19.15
In this paper, a novel road sign detection method is
proposed. The method first uses color information to
localize potential road signs and then, identifies the
correct shape of the detected signs. Shape detection
is based on shape representation using Gielis curves
which provides an elegant way to handle all common
road sign types, i.e. triangles, rectangles, octagons,
and ellipses. Experimental results show the robust-
ness of the approach in detecting traffic signs of var-
ious shapes. The method is invariant to in-plane ro-
tation and to small perspective distortion due to the
introduction of a rotational offset as a parameter in
the fitting algorithm. It is also able to detect signs
of different sizes in the image. The different causes
of failure can be considered by improving the color
segmentation method robustness to illumination ef-
fect. We could for example apply a color constancy
algorithm prior to segmentation. Another improve-
ment would be the introduction of a parameter in the
fitting algorithm to account for strong perspective dis-
tortions. Eventually, the general least square formu-
lation of the problem could be replaced by a more ro-
bust version, using M-estimators for instance, in or-
der to better handle outliers to improve the results in
presence of degenerate contours due to occlusions or
strong light variations.
Ballard, D. H. (1981). Generalizing the hough trans-
form to detect arbitrary shapes. Pattern Recognition,
Bascon, S. M., Arroyo, S. L., Jimenez, P. G., Moreno, H. G.,
and Ferreras, F. L. (2007). Road Sign Detection and
Recognition Based on Support Vector Machines. In
IEEE Transactions on Intelligent Transportation Sys-
tem, volume 8, pages 264–278.
Escalera, A., Armingol, J. M., and Mata, M. (2003). Traffic
Sign Recognition and Analysis for Intelligent Vehi-
cles. Image and Vision Computing, 21:247–258.
Escalera, A., Armingol, J. M., Pastor, J. M., and Rodriguez,
F. J. (2004). Visual Sign Information Extraction and
Identification by Deformable Models for Intelligent
Vehicles. In IEEE Transactions on Intelligent Trans-
portation Systems, volume 5, pages 57–68.
Fang, C. Y., Chen, S. W., and Fuh, C. S. (2003). Road Sign
Detection and Tracking. In IEEE Transactions on Ve-
hicular Technology, volume 52, pages 1329–1341.
Fleyeh, H. (2004). Color Detection and Segmentation for
Road and Traffic Signs. In Proceedings of the 2004
IEEE Conference on Cybernetics and Intelligent Sys-
tems, pages 808–813, Singapore.
Fougerolle, Y. D., Gribok, A., Foufou, S., Truchetet, F.,
and Abidi, M. A. (2005). Boolean Operations with
Implicit and Parametric Representations of Primitives
using R-functions. In IEEE Transactions on Visualisa-
tion and Computer Graphics, volume 11, pages 529–
Gao, X., Podladchikova, L., Shaposhnikov, D., Hong, K.,
and Shevtsova, N. (2006). Recognition of traffic signs
based on their colour and shape features extracted us-
ing human vision models. Journal of Visual Commu-
nication and Image Representation, 17(4):675–685.
Gavrila, D. (1999). Traffic sign recognition revisited. In
GAGM-Symposium, pages 86–93.
Gielis, J. (2003). A Generic Geometric Transformation that
Unifies a Wide Range of Natural and Abstract Shapes.
American Journal Botany, 90:333–338.
Loy, G. and Barnes, N. (2004). Fast shape-based road sign
detection for driver assistance system. In Proceedings
of International Conference on Intelligent Robots and
Systems (IROS 04), pages 70–75.
Loy, G. and Zelinsky, A. (2002). A fast radial symmetry
transform for detecting points of interest. In Proceed-
ings of 7th European Conference on Computer Vision,
pages 358–368.
Nguwi, Y. Y. and Kouzani, A. Z. (2008). Detection and
Classification of Road Signs in Natural Environment.
In Neural Compututing and Application, pages 265–
289. Springer-Verlag London Limited 2007.
Paclik, P., Novovicova, J., Pudil, P., and Somol, P. (2000).
Road sign classiffication using laplace kernel classi-
fier. Pattern Recognition Letters, 21:1165–1173.
Varun, S., Singh, S., Kunte, R. S., Samuel, R. D. S., and
Philip, B. (2007). A road traffic signal recognition sys-
tem based on template matching employing tree clas-
sifier. In Proceedings of International Conference on
Computational Intelligence and Multimedia Applica-
tions (ICCIMA 07), pages 360–365, Washington, DC,
VISAPP 2012 - International Conference on Computer Vision Theory and Applications