Circular Traffic Signs Recognition Using The Number of Peaks Algorithm

Khaled M. Almustafa
International Journal of Image Processing (IJIP), Volume (8) : Issue (6) : 2014 514
Circular Traffic Signs Recognition Using The Number of Peaks
Algorithm
Khaled M. Almustafa kalmustafa@psu.edu.sa
Prince Sultan University.
P.O.Box No. 66833
Riyadh 11586, K.S.A
Abstract
Smart cars nowadays include embedded computers to guide the driver in his trip. An important
application that should be added to any car is the detection and recognition of traffic signs. In this
paper, we focus on the recognition of a wide set of circular traffic signs using the Number of
Peaks Algorithm [1]. After detecting a traffic sign, the algorithm draws three horizontal lines and
three vertical lines across the image. The number of peaks (crossing from a black pixel to a white
pixel) is calculated for each of the six lines as the image is scanned from right to left (for
horizontal lines) or top to bottom (for vertical lines). The resulting numbers of peaks are used by
the decision-tree-like search algorithm to distinguish between 51 circular road signs with a mean
detection time of 8 milliseconds, 100% detection rate and in a fairly noisy environment.
Keywords: Traffic Signs Recognition, Pattern Recognition, Image Processing, Autonomous
Cars.
1. INTRODUCTION
Automatic detection and recognition of traffic signs is an important addition to any smart car.
Having such a tool in a car would alert the driver to possible obstacles or changes in the road and
therefore reduce the possibility of accidents. Previous systems for real time detection of traffic
signs are limited to a small set of circular or triangular signs [2, 3]. A system capable of
recognizing all traffic signs within an acceptable amount of time for a moving car is highly
desirable.
In this paper, a system is described that recognizes a wide set of circular traffic signs. The system
uses a novel algorithm, the number of peaks algorithm, to differentiate between 51 circular road
signs in a fast and reliable manner. Many researchers studied the detection and recognition of
traffic signs since 1996 [1-21]. Most of the approaches to traffic sign recognition follow a two-step
algorithm. The first step is the detection of a traffic sign in the image and the second step is to
recognize the sign. The second step may involve classification of the sign first into predefined
classes, such as triangular signs or speed signs, followed by recognition of a specific sign [4, 5,
6]. In the recognition phase, various approaches are available in the literature [7]. Most
approaches are pixel-based using cross-correlation template matching [4, 8] or neural networks
[5]. However, other approaches exist that are feature-based. For example, in [9] statistical
properties, such as moments found from the binary images of the central part of sign candidates,
were used. Local edge orientations and density at arbitrary fixation points were used [10].
A different approach was also presented in [6, 7] based on the similarities of the detected sign
and the sign images stored in the template. Equiangular polygons are detected in a filtered image
then a discrete-color image of this object is compared with model images. Another technique [11]
uses error correcting output codes (ECOC) to build a system for multi-class classification of traffic
signs. In all cases classification methods consider a limited number of signs such as six circular

Khaled M. Almustafa
signs [2] or blue traffic signs [3] and are characterized by varying recognition rate 81% [3], 76% to
91% [7], and 98.66% [11].
Section 2 introduces the number of peaks algorithm. Section 3 introduces the decision trees used
in the algorithm and details how a specific sign is distinguished based on number of peaks in
horizontal and vertical lines. Section 4 discusses the performance of the algorithm. Section 5
concludes the paper and gives suggestions for future research.
2. THE NUMBER OF PEAKS ALGORITHM
The number of peaks algorithm works after a traffic sign is detected. Figure 1 shows all possible
images of circular signs recognized by the Number of Peaks Algorithm. Figure 1 shows each sign
along with a number assigned to it. This number is used in the flowcharts of the algorithm to
distinguish between the signs as shown in the next section.
FIGURE 1: Circular Traffic signs recognized by the Number of Peaks Algorithm and their number
representation.
Once a traffic sign is detected, its negative image representation is generated as shown in Figure
2.
FIGURE 2: A sign and its negative image representation.
Then a factor f is selected from a discrete set {1/3, 1/4, 1/5, 1/9, 1/10}. This factor is used to
select the positions of three horizontal lines and three vertical lines to be drawn across the image.
Lines T, H and B (in Figure 3) are horizontal lines. Line T is at a distance f of the total image
length taken from the image top. Line H is drawn at the middle of the image horizontally. Line B is

Khaled M. Almustafa
drawn at a distance f of the total image length taken from the image bottom. Similarly, lines R, V
and L are three vertical lines drawn across the image at distances f of the total width from the
image right boundary, at the middle and f of the total width from the image left boundary
respectively.
FIGURE 3: Crossing lines of an Image.
F=1/3 H T B V L R
26 2 3 2 2 1 1
50 1 1 1 1 1 1
21 2 1 1 2 2 2
24 2 1 2 2 2 2
25 2 1 2 2 2 4
28 2 2 5 2 2 2
30 2 2 2 2 2 2
31 2 3 2 2 2 2
23 3 2 3 2 2 3
32 3 2 3 3 2 2 R 1/5 (1)
33 3 2 3 3 2 2 R 1/5 (2)
27 3 3 3 2 2 2
51 3 3 2 3 2 2
29 4 3 2 2 2 4
37 4 3 5 4 2 2
18 6 4 4 3 2 4
20 4 4 2 2 3 3
10 4 5 3 3 3 4
16 5 3 3 3 3 3
19 4 4 3 2 3 3
4 4 4 4 3 3 4
5 4 4 4 4 3 4
39 3 4 4 5 3 3
45 4 4 4 3 3 3
35 10 4 5 5 3 6
13 6 2 2 4 4 2
6 5 4 7 4 4 3
8 3 4 5 4 4 3
34 3 3 4 3 4 3
9 4 3 3 2 4 4
42 4 3 3 4 4 4
7 5 4 5 3 4 4
2 5 3 4 4 4 4 R 1/5 (2)
38 5 3 4 4 4 4 R 1/5 (3)
17 6 4 3 5 4 4
36 6 3 5 2 4 4
41 4 3 2 3 4 5
40 6 2 2 2 5 5
1 7 4 5 4 5 6
3 3 4 5 4 5 4
43 4 3 5 4 5 4
14 5 3 6 5 5 4
15 2 4 6 4 5 4
46 7 5 7 7 5 5
47 8 4 7 6 5 7
11 4 4 3 6 6 4
12 8 4 4 5 6 4
22 4 4 5 4 6 3
44 3 9 5 6 6 5
48 8 5 7 7 6 6
49 9 7 5 7 7 6
TABLE 1: Grouping of the Signs by the number of Peaks of the L line.
Factor= 1/3 of the total length of the
image

Khaled M. Almustafa
A Peak is defined as a crossing from a black pixel to a white pixel [1, 22] as the image is scanned
from right to left (for lines T, H and B) or top to bottom (for lines R, V and L). A line starting with
white pixels is considered to have one peak at the beginning. Line T in Figure 3 contains two
sections where there is a transition from black pixels to white pixels. This means that for the line
labeled H, the algorithm would return a value of 2.
3. DECISION TREES FOR THE NUMBER OF PEAKS ALGORITHM
The number of peaks for all 6 lines using factors 1/3 and 1/5 were calculated for all the circular
traffic signs considered (51 in total). Those numbers were studied to find out the minimum
number of lines that can be used to recognize each image. For all the images, we started out with
the line that would divide those images into the maximum number of groups. Then use another
line to keep dividing the groups until recognizing all the images. The results are summarized in
Table 1. Table 1 is sorted according to the lines that were used to identify the signs. Given a
circular traffic sign to be recognized, the algorithm starts with a factor of 1/3. The number of
peaks for the L line is found. This number divides all the circular traffic signs into seven groups as
shown in Table 1. If this number is 1, then the studied sign is either number 26 or number 50.
The number of peaks for the V line is then found. If this value is 2 then the sign is number 26. If
this value is 1, then the sign is number 50.
These results can also be seen in the flowchart shown in Figure 4. If the number of peaks for the
L line is 3, then the candidate sign is a sign belonging to the set {20, 10, 16, 19, 4, 5, 39, 45, 35}.
The number of peaks for additional lines should be found ( B then T then R then V) to identify
those signs. The remaining parts of the flowchart for the nodes labeled H, R1 and B2 are
continued in Figures 5, 6 and 7 because of space limitation.
FIGURE 4: Flowchart for the First test of the algorithm.

Khaled M. Almustafa
FIGURE 5: Flowchart for node H of Figure 4.
FIGURE 6: Flowchart for node R1 of Figure 4.

Khaled M. Almustafa
FIGURE 7: Flowchart for node B2 of Figure 4.
4. RESULTS
The algorithm was repeated 1000 times to recognize each of the 51 signs on an HP 8200 Elite
with an I5 2500 CPU at 3.30 GHz. The average detection time for a sign is 8 milliseconds.
In order to test the performance of the Number of Peaks Algorithm in blurred images, noise was
artificially introduced to the signs considered and the algorithm was run on the resulting noisy
images. Artificially introduced errors were simulated using equation (1)
P = Q + µN (1)
Where matrix Q is representing the image, and N is of a same size as Q, which consists of
uniformly distributed pseudo-random numbers and µ is a constant, which is used to control the
amount of noise to introduce in the image. Performance is measured using the Frobenius norm of
(P - Q). The Frobenius norm of an m × n matrix A is defined by the following equation (2).
(2)
A graph of the mentioned norm as a function of µ is shown in Figure 8. We can see that the solid
line shows the averaged calculated threshold over the range of values of µ for different type of
circular signs where the noisy image can be recognized by the algorithm, and the dashed line
shows the averaged threshold for the value of µ where the character would not be recognized by
the Human eye. The human eye threshold was determined based on our responses to the given
noisy images. It turns out that the value of the threshold of recognition is equal to µ = 1.08, with
an equivalent human eye threshold of 1.7. This means that our system would be able to work
within a fairly noisy environment even when the amount of noise changes the features important
to the algorithm drastically. However the system is still inferior to a human being in the same
environment.

Khaled M. Almustafa
FIGURE 8: Performance of the Frobenius norm.
Even in artificially induced noisy images, the detection rate of the Number of Peaks algorithm is
100% up to some level of noise added.
FIGURE 9: A sign with added noise having µ=1.08.

Khaled M. Almustafa
Figure 9 shows in the top row an image used in this simulation and its negative generated image.
In the second row, the noisy image having a µ=1.08 is shown and the corresponding negative
image that was successfully recognized by the number of peaks algorithm.
5. CONCLUSION AND FUTURE RESEARCH
In this paper an algorithm for recognition of 51 circular road signs was presented. The algorithm
uses a simple method that requires a minimum number of image processing combined with a
decision-tree-like search algorithm. The algorithm was repeated 1000 times to recognize each of
the 52 signs on an HP 8200 Elite with an I5 2500 CPU at 3.30 GHz. The average detection time
for a sign ranged between 8 milliseconds for sign number 11 to 9.5 milliseconds for sign number
23, with an average detection time of 8.47 milliseconds. Even in artificially induced noisy images,
the detection rate of the Number of Peaks algorithm is 100% up to some value of the noise
parameter µ.
The next step of this research is to extend this sign recognition algorithm to recognize signs of
any shape using the same principle. Flowcharts similar to the ones presented in Figures 4 to 7
should be drawn to detect all traffic signs. The system should then be integrated into a full real
time system to detect and then recognize all classes of traffic signs in any country. Shape
recognition can be investigated to help detect and locate the sign in real time images. The
algorithm should be implemented on portable hardware and installed in smart cars. Since the
time to detect and recognize a sign should be minimized for the system to be useful in high speed
zones, a parallel version of this algorithm should be created and tested on multicore processors
and compared to the performance of implementing this algorithm on Field Programmable Gate
Arrays (FPGA).
6. ACKNOWLEDGMENT
The authors would like to extend their sincere thanks to Prince Sultan University (PSU), Riyadh,
K.S.A. for its support to the project.
7. REFERENCES
[1] L. Hamandi, K. AlMustafa, R. Zantout, H. Obeid, Recognition of Triangular Traffic Signs
Using the Number of Peaks Algorithm, accepted for publication in The 2nd International
Conference On Advances In Computational Tools For Engineering Applications, Dec. 12-
15, 2012, Beirut, Lebanon.
[2] K. Baba, Y. Hirai, Real-time recognition of traffic signs using opponent color, Proceedings
of the 14th World Congress on Intelligent Transport Systems (ITS) 2007, pp. 5127-5138.
[3] M. H. Alsibai, Y. Hirai , Real-Time Recognition of Blue Traffic Signs Designating Directions,
International Journal of Intelligent Transportation Systems Research, May 2010, Volume
8, Issue 2, pp 96-105.
[4] G. Piccioli, E. De Micheli, P. Parodi, M. Campani, A robust method for road sign detection
and recognition, Image and Vision Computing 14 (3) (1996) pp. 209–223.
[5] A. de la Escalera, L.E. Moreno, M.A. Salichs, J.M. Armingol, Road traffic sign detection and
classification, IEEE Transactions on Industrial Electronics 44 (6) (1997) pp. 848–859.
[6] P. Paclík, J. Novovicova, P. Pudil, P. Somol, Road signs classification using the Laplace
kernel classifier, Pattern Recognition Letters 21 (13–14) (2000) pp. 1165–1173.
[7] A. Ruta, Yongmin Li,Xiaohui Liu./ Real-time traffic sign recognition from video by class-
specific discriminative features, Pattern Recognition 43(2010) pp. 416—430.

Khaled M. Almustafa
[8] A. de la Escalera, J.M. Armingol, J.M. Pastor, F.J. Rodríguez, Visual sign information
extraction and identification by deformable models for intelligent vehicles, IEEE Transactions
on Intelligent Transportation Systems 5 (2) (2004) pp. 57–68.
[9] P. Paclík, J. Novovicová, R.P.W. Duin, Building road-sign classifiers using a trainable
similarity measure, IEEE Transactions on Intelligent Transportation Systems 7 (3) (2006) pp.
309–321.
[10]X.W. Gao, L. Podladchikova, D. Shaposhnikov, K. Hong, N. Shevtsova, Recognition of traffic
signs based on their colour and shape features extracted using human vision models, Journal
of Visual Communication and Image Representation 17 (4) (2006) pp. 675–685.
[11]Escalera et. al. Traffic sign recognition system with β-correction Machine Vision and
Applications (2010) 21, pp. 99–111.
[12]Y. Aoyagi, T. Asakura, A study on traffic sign recognition in scene image using genetic
algorithms and neural networks, Proceedings of the 1996 IEEE IECON 22nd International
Conference on Industrial Electronics, Control, and Instrumentation, vol. 3, 1996, pp. 1838–
1843.
[13]C. Bahlmann, Y. Zhu, V. Ramesh, M. Pellkofer, T. Koehler, A system for traffic sign detection,
tracking and recognition using color, shape, and motion information, in: Proceedings of the
IEEE Intelligent Vehicles Symposium, 2005, pp. 255–260.
[14]P. Douville, Real-time classification of traffic signs, Real-Time Imaging 6 (3) (2000) pp. 185–
193.
[15]C.-Y. Fang, S.-W. Chen, C.-S. Fuh, Roadsign detection and tracking, IEEE Transactions on
Vehicular Technology 52 (5) (2003) pp. 1329–1341.
[16]H. Fleyeh, Color detection and segmentation for road and traffic signs, in: Proceedings of the
IEEE Conference on Cybernetics and Intelligent Systems, Singapore, vol. 2, 2004, pp. 809–
814.
[17]M.A. Garcia-Garrido, M.A. Sotelo, E. Martin-Gorostiza, Fast traffic sign detection and
recognition under changing lighting conditions, in: Proceedings of the IEEE International
Conference on Intelligent Transportation Systems, 2006, pp. 811–816.
[18]Hatzidimos, Automatic Traffic Sign Recognition in Digital Images, Proceedings of the
International Conference on Theory and Applications of Mathematics and Informatics -
ICTAMI 2004, Thessaloniki, Greece, pp. 174 – 184.
[19]J. Miura, T. Kanda, Y. Shirai, An active vision system for real-time traffic sign recognition, in:
Proceedings of the IEEE Conference on Intelligent Transportation Systems, Darborn, MI,
USA, 2000, pp. 52–57.
[20]A. Ruta, Y. Li, X. Liu, Towards real-time traffic sign recognition by class-specific
discriminative features, in: Proceedings of the 18th British Machine Vision Conference,
Coventry, United Kingdom, vol. 1, 2007, pp. 399–408.
[21]A. Ruta, Y. Li, X. Liu, Traffic sign recognition using discriminative local features, in:
Proceedings of the 7th International Symposium on Intelligent Data Analysis, Ljubljana,
Slovenia, 2007, pp. 355–366.
[22]K. AlMustafa, R. Zantout, H. Obeid, “Peak Position, Recognizing Characters in Saudi License
Plates”, 2011 IEEE GCC Conference and Exhibition for Sustainable Ubiquitous Technology,
Dubai, United Arab Emirates, February 19-22, 2011, pp. 186-189.

Circular Traffic Signs Recognition Using The Number of Peaks Algorithm

More Related Content

What's hot (20)

Viewers also liked (17)

Similar to Circular Traffic Signs Recognition Using The Number of Peaks Algorithm (20)

Recently uploaded (20)

Circular Traffic Signs Recognition Using The Number of Peaks Algorithm