7A-5-2 
Although there is an extensive literature covering 
traffic sign detectionFrecognition and localization with 
potential applications in intelligent transportation sys- 
temrdevelopment of a practical automated traffic sign 
recognition system remains a challenge. Many factors 
affect traffic sign detection. Rust and faded paintr 
sign bendingr weather conditions (light or overcast)T 
shadowrocclusionsrreflection of sunlight motionrand 
vibrations of the van affect image quality. And within 
an imagerif the size of a sign is too smalirexisting vi 
sion algorithms may treat the sign as noise. While 
many techniques for color segmentation have been 
suggested!? segmentation remains difficult due to the 
abundance of colors in traffic scenes and the influence 
of weatherHighting conditions and shadows. 
Most developed algorithms detect signs from image 
sequences using color segmentation and shape extrac- 
tionrthat isl?edge detectionFHough transformPor tem 
plate matching. They then match extracted features 
with traffic sign models to recognize and identify road 
signs. 
Traffic signs consist mainly of artificial pure colors: 
redT oranger yellowr brownr greenr black and white. 
We therefore propose a fuzzy color segmentation algo 
rithm to extract these artificial colors. In one traffic 
sign libraryPabout one-sixth are black and whiterfor 
instancer a white background with black text or ar 
rows or a black background with white text or arrows. 
To detect white and black traffic signsTone visual cue 
is shape information and another is text information. 
We make use of text regions (the high contrast be 
tween foreground symbols and a panel background) 
to reduce the search space for black and white sign 
detection. Moreoverrtraffic signs are usually mounted 
on poles or lamp posts. This suggests searching for 
vertical poles or lines and then searching for possible 
traffic signs on them. 
In the next section 3Tan architecture for the auto 
mated traffic signs recognition system for mobile map 
ping is depicted. In section 4Ta color segmentation al 
gorithm and a blob analysis algorithm are presented. 
In section 5rshape analysis based on Fourier descrip 
tors is made. In section 6Tan algorithm for text re 
gions extraction based on texture energy is discussed. 
3 Overview of the traffics sign 
recognition system 
The traffic sign detectionTpredicationlYecognition and 
location system depicted in Figure 2 consists of four 
modules: (i) sign detectionr (ii) sign predictionT (iii) 
Figure 2: A traffic sign recognition system 
sign recognitionTand (iv) sign location. This article 
describes an implementation of the first one. 
4 Color segmentation algorithm 
To overcome difficulties of designing a general color 
segmentation algorithm [6]r we give a special algo 
rithm to extract color regions from an image. As noted 
earlierrtraffic signs are mainly composed of artificial 
redT orangeTyellowr greenT and blue colors. In con 
trast to segmenting a natural scene with millions of 
possible colorsPwe need only focus on artificial colors 
to improve traffic sign detection. 
4.1 Color space 
A color can be defined as a mixture of tristimulus 
components. Many color representation spaces are 
in use todayrsuch asTRGBTHSLTand La*b*. Be 
cause of lighting conditionsrsigns under shadows are 
different from non-shadowed ones. Normalized RGB 
color space partially overcomes this drawback. It does 
not give an invariant uniform shifting of RGB compo 
nents. Other color spaces are often considered. They 
consist of chromatic components (hue and saturation) 
and an intensity component. Chromatic components 
are noise sensitive. For exampleTin the HSL color 
spaceTthe H and S values of black and white traffic 
signs are very sensitive to lighting conditions.