Figure 1 : Our 3 steps strategy. 
(0 + 
G-B 
MAX-MIN 
B-R 
) x 60 
V+ MAX-min) XW 
✓ . D r-' 
if R 
if G 
MAX, 
MAX, 
(4 + 
Q 
MAX-MIN 
) X 60 if B = MAX, 
MAX - MIN 
MAX 
V = MAX, 
where: 
MAX — max(R, G, B) 
MIN = min(R, G, B) 
0) 
4 CIRCULAR SIGN DETECTION 
The shape detection have to detect all the types of road 
signs (the rectangular, triangular and circular road signs). 
In this first version of work we choose to focus on the cir 
cular road signs because they are the most common. The- 
orically, a circle appears as an ellipse in perspective im 
ages. The quantity of perspective deformation depends on 
the angle between image and the circle plane. Often, road 
signs belong to a traffic lane and supposed to provide infor 
mation to drivers in the same lane. In this case perspective 
deformation is negligible. This is the reason why most of 
the Driver Assistance Systems (ADAS) ignore perspective 
deformation. 
We aim at extracting all visible road signs within an im 
age what ever their orientation is. This is interesting in 
both database generation and the use of road signs as vi 
sual landmarks for positioning purposes. Thus, an ellipse 
detection algorithm is investigated (Section 4.1). 
Figure 2: Color detection results, a) our running example 
RGB image, b) blue color mask, c) labeling independent 
connected pixels. 
4.1 Ellipse Detection 
Input of this step is a set of image windows provided by 
the color detection step. We use edge points for ellipse de 
tection. In each image window, edges are extracted using 
Canny-Deriche edge detector (Deriche, 1987). 
An ellipse is defined with five parameters (2 for the center, 
2 for the axes length and one for orientation). Equation 2 
express equation of ellipse. In this Equation p and q stand 
for ellipse center. Orientation and axes length depend on 
a, b and c. 
a{x - p) 2 -1- 2b(x - p)(y -q) + c{y - q) 2 = 1 (2) 
This equation is not linear. We make use the Pascal's the