2023 International Conference on Sustainable Emerging Innovations in Engineering and Technology (ICSEIET)

Traffic Signal Recognition System using Deep

Learning
2023 International Conference on Sustainable Emerging Innovations in Engineering and Technology (ICSEIET) | 979-8-3503-2919-3/23/$31.00 ©2023 IEEE | DOI: 10.1109/ICSEIET58677.2023.10303609

5DPHVK1XWKDNNL $EGXOODK$DPHHQ 1DZQLW.XPDU

';&7HFKQRORJ\,QGLD 6UL*XUX7HJK%DKDGXU
$WULD,QVWLWXWHRI7HFKQRORJ\ DEGXOODKDDPHHQ#G[FFRP
%HQJDOXUX,QGLD ,QVWLWXWHRI0DQDJHPHQW 
QXWKDNNLUDPHVK#DWULDHGX ,71HZ'HOKL,QGLD
6KDODEK.0LVKUD
QDZQLWNXPDU#JPDLOFRP
%KDUDWL9LG\DSHHWKV¶&ROOHJHRI(QJLQHHULQJ
1HZ'HOKL,QGLD
VKDODEKPLVKUD#JPDLOFRP
Abstract— The intelligent traffic system relies heavily on traffic recognising traffic signs. According to a British study,
sign recognition. It can decrease driver fatigue and improve Intelligent speed adaption (ISA) might cut mortality in half.
traffic safety. The traffic sign recognition system (TSRS) is a key The benefits of ISA, according to the RTA (NSW Australia)
element of the intelligent transportation system (ITS). Accurately ISA experimental outcomes are: -Increased speed zone
identifying traffic signs can improve driving safety. This paper
compliancy as well as reduction in the extent and continuance
proposes a traffic sign recognition technique formulated on deep
learning, taking into consideration the main objective of of speeding.
identifying as well as categorizing circular and spherical signs. In The actual and discerned advantages of ISA include lower
conclusion, deep learning is utilized in classifying the identified risks in accidents , lesser noise and emptying out of pollutants
road traffic signs. The paper concludes with a focus on the [21-24]. If vehicles and pedestrians do not pay attention to this
proposed system technology, that also utilizes neural networks. information, traffic accidents may occur. In addition to the
As a result, traffic-sign recognition stands as both demanding as- rising need for vehicle intellect, it is crucial to utilize
well-as appealing subject, and an important area of study within computer technologies to detect and recognise traffic signs
the field of traffic engineering research. automatically.

Keywords— Artificial Intelligence, Automatic recognition,

Computer vision, Deep Learning, Genetic Algorithm.

I. INTRODUCTION

As stated by WHO, over 1.3 million people are being killed

through road accidents every year worldwide. Also, around
20-50 million people are being affected with non-fatal injuries,
most of them sustaining some form of disability as a result of
their injury. Individuals, their families, and nations as a whole
suffer enormous economic losses as a result of road traffic
injuries. Everyday vehicle collisions are caused primarily by
poorly visible road signs. Driving can be a difficult task for
those with colour vision deficiencies, so it is critical to have a
system that can assist drivers in driving safely. The traffic sign
recognition system is essential in order to control traffic and
collision avoidance. It can warn distracted drivers about
changes in approaching lights or even trigger an automatic
response from the vehicle itself.
The traffic sign recognition acts as a multicategory
classifying challenge that occurs because of the imbalance in
class frequencies. It has been a demanding real-life issue of
computer vision accompanied by several pragmatic
applications that turned out to be the subject of
many researches for decades. On this subject, numerous
researches have been published. Basically, the expert systems
bank on traffic sign identification, classification as well as
detection namely traffic assistance along with automated Fig. 1. Block diagram of a generalized AI based System.
driving systems. It immediately aids human and automated
driving systems in successfully detecting as well as

979-8-3503-2919-3/23/$31.00 ©2023 IEEE

636
Authorized licensed use limited to: Consortium - Algeria (CERIST). Downloaded on November 20,2023 at 20:44:27 UTC from IEEE Xplore. Restrictions apply.
 2023 International Conference on Sustainable Emerging Innovations in Engineering and Technology (ICSEIET)

Accurate traffic signal recognition on public roads is a B. Fuzzy Theory

critical step for any automated driving system. Numerous
works in computer vision have been undertaken in order to Fuzzy theory is a technique of dealing with uncertain
increase safety on roads and significantly lessen the risk of information and is established as a way of presenting
accidents. It has been divided mainly into two types of measures that are mathematical to an extensive range of
technologies namely traffic-sign detection and recognition. confusing occurrences. Gang-Yi Jiang et al. [13] introduced a
The essential traits that helps drivers to get the information on new method for detecting a turning highway by recognising
roads are shape and colour and traffic signs carries certain traffic signing, where fuzzy logic has been employed for
consistent properties which can be utilized in detecting and improving colour images by not affecting their colouring.
classifying. Each country's traffic sign colours are nearly same, Also, a neural network is used for the classification of traffic
typically comprising of plain colours such as red, yellow, blue signs due to its excellent adaptability in case of noisy
and so on .Also having standardized shapes such as circles, situations. Because this theory is well suited in explaining the
rectangles, etc. External elements, such as weather, have a phenomenon of ambiguity as well as unstable issues.
significant influence on the appearance of traffic signs. Hypothetical outcomes indicate that the proposed algorithm is
Recognizing traffic lights accurately and dependably in all remarkably easy and reliable, it also executes relevantly in
time and weather conditions is quite difficult. real-world object detecting.

II. STATE-OF-THE-ART

A. Linear and Non-linear SVM

In order to recognise and detect traffic signs out of image
series, Kiran C. G. et al. [20] in his work utilized a colour
intelligence. For detecting the traffic signs, colour based
segmentation algorithms have been used for traffic. They
employed a product with increased hue and saturation
components to boost segmentation performance. Linear
support vector machines were used along with the divided
blobs' Distance to Border information to improve shape
classification performance. Multi-classifier non-linear support
vector machines containing the edge related pixels of interest
being the characteristic are used for recognising traffic signs.
This recognition system generally comprises of three phases:-
1. Colour Segmentation: By thresholding the input image,
blob candidates are retrieved out of the input test image.
Chromatic signs are segmented using HSI chromaticity
thresholding, while achromatic signals are segmented utilizing
the process of achromatic decomposition.
2. Shape Classification: Blobs produced through the colour
segmentation procedure are analysed using linear SVM in
accordance with their shape in this stage.
Fig. 3. Boolean and Fuzzy logic
3. Pattern Recognition: Non-linear SVM is being used to
recognize the shape categorized blobs depending on their
patterns.
C. Dynamic Pixel Aggregation Technique

S. Vitabile et al. [19] describe a solution for the entire

traffic sign recognition as well as detection challenge that
improves the image segmentation results. On the HSV colour
space, the process of segmenting colour is done through
initiating a dynamic approach within the pixel augmentation
procedure. When the processed images contains more sign
pixels having coordinates within the HSV achromatic and/or
unsteady zones, system performance suffers. The A (RS)2
provides good detection and identification rates in a variety of
light conditions. The dynamic threshold reduces hue
instability in real-world scenes that are affected by external
Fig. 2. Linear vs Non-linear Problem brightness variations.

637
Authorized licensed use limited to: Consortium - Algeria (CERIST). Downloaded on November 20,2023 at 20:44:27 UTC from IEEE Xplore. Restrictions apply.
 2023 International Conference on Sustainable Emerging Innovations in Engineering and Technology (ICSEIET)

D. Fusion using Belief Theory likelihood cross-validation approach is used for optimizing
smoothing parameters of the Laplace kernel. The EM
J. Ph. Lauffenburger et al. [18] developed an improved algorithm-based maximisation of with regard to cross-
vehicle localization approach based on a multi-criteria fusion validated log-likelihood function was indeed a key advantage
solution that allows for the evaluation of the driving of this technique, as it allowed for quick computation and
circumstance identification process' confidence level. Between efficient learning.
the data provided by each source, fusion utilising Belief
Theory is undertaken. To determine the proper driving G. Posterior Classifier Combination
circumstances with respect to speed, the union combines
many parameters from the navigation system's digital map It is critical to develop a robust and reliable object
along with legal speed identified through the speed sign recognition technique that bases decision making on
refining network. The usefulness of this strategy is uncertainty calculi such as probabilistic reasoning in this
demonstrated by real-world outcomes through a driving research by Lucas Paletta [15] for vision-based mobile
circumstance where the two sources get into dispute. mapping of railway infrastructure. In a probabilistic context, a
Radial Basis functions (RBF) neural network is described
E. Local Linear Embedding which exhibits appearance-based object recognition of road
signals and railways signboards. By the comparison of various
By utilizing general purpose hardware along with low cost classifier combination strategies reveals that a classifier
sensors, Xiaolong Teng et al. [17] suggested a real-time vision prioritising system based on an information theoretic selection
framework across the visual interaction situations by hand criterion yields the best recognition results.
gesture detection. By utilizing the local linear embedding
algorithm, the feature extraction procedure is the most H. Genetic Algorithm
significant aspect of the recognition approach. It is a rapid
detection algorithm that can handle a high amounts of hand Toshiyuki Asakura et al. [14] describe a method for
signalling in resistance to various internal backgrounds along recognising and detecting the positioning regarding a circle
with the lightening conditions while obtaining the significant traffic sign through a moving picture using a genetic
hand region from the entire image. The system's identification algorithm (abbreviated as GA). The location and measure
rates show that it is resistant to comparable postures. relating to traffic sign are both considered genetic data. The
correct traffic sign circular design has been chosen from a vast
variety of circular patterns in the backdrop, and then a circle
standard pattern is created for pattern matching on the image.
In this study, a searchable GA is used to solve this challenge,
moreover the circle pattern within the image is equalled with
the help of genetic data. SVF is used for detecting the position
of a traffic sign in real time. SVF may then extract a specified
colour while quickly removing the contour. The method,
which limited the search zone, in real time could be used to
check the position of a traffic sign and therefore this SVF-
based recognition system is effective.

Fig. 4. Local Linear Embedding Pipeline

F. Laplace kernel classifier

The main purpose of this study work by P. Paclik et al. [16]

was to demonstrate the behaviour concerning the Laplace
kernel classifier in relation to a reallife situation just as traffic
sign identification besides comparing it to various approaches.
The Laplace probability density was used to construct a novel
kernel rule for traffic sign classification. The pseudo Fig. 5. Genetic Vs Neural Network

638
Authorized licensed use limited to: Consortium - Algeria (CERIST). Downloaded on November 20,2023 at 20:44:27 UTC from IEEE Xplore. Restrictions apply.
 2023 International Conference on Sustainable Emerging Innovations in Engineering and Technology (ICSEIET)

III. FUTURE SCOPE [10] Yongping Wang, Meiping Shi, Tao Wu, “A Method of Fast and Robust
for Traffic Sign Recognition”, 2009 Fifth International Conference on
In the contemporary state-of-the-art, there is zero human- Image and Graphics, ISBN:978-1-4244-5237-8; DOI:
10.1109/ICIG.2009.130
machine interface, that could be a focus of upcoming [11] Jyoti Kumari, R. Rajesh, K. M. Pooja “Facial Expression Recognition:
researches for the researchers because it is important for an A Survey”,
effective ADAS system. The availability of public databases [12] Omar Belghaoutia, Wahida Handouzia, Mohamed Tabaab, Improved
is another key issue that must be addressed. Although certain Traffic Sign Recognition Using Deep ConvNet Architecture, (2020)
[13] Gang-Yi Jiang, Tae-Young Choi, and Suk-Kyo Hong, Turning
databases are available to the public, they are rarely used. A Roadway Detection Based on Sign Information, (1999)
combinational technique could be employed in future research [14] T. Asakura, Y. Aoyagi, O.K. Hirose ,”Real-time recognition of road
to lower the dependability factor, as well as other challenges, traffic sign in moving scene image using new image filter”, SICE 2000
in real-time applications. Furthermore, traffic signs can be Proceedings of the 39th SICE Annual Conference., ISBN:0-7803-
9805-X; DOI: 10.1109/SICE.2000.889645
represented by GPS or RFID for accurate recognition. This [15] Lucas Paletta ,Detection of traffic signs using posterior classifier
can be utilized in a wide range of vehicles. It is possible to combination, (2002)
make this kind of technology more affordable for the common [16] P.PaclÕғk, J.Novoviþová, P.Pudil, P.Somol, Road sign classification
people while keeping the vehicles eco-friendly. The size of the using Laplace kernel classifier, (2000)
[17] Xiaolong Teng, Bian Wu, Weiwei Yu, Chongqing Liu, A hand gesture
devices used for this purpose, such as cameras, sensors and recognition system based on local linear embedding, (2005)
other equipment, may be reduced for a better experience. [18] J. Ph. Lauffenburger, B. Bradai, M. Basset, F. Nashashibi, Navigation
and Speed Signs Recognition Fusion for Enhanced Vehicle Location,
IV. CONCLUSION (2008)
[19] S . Vitabile, G. Pollaccia, G. Pilato and E Sorbello, “ROAD SIGNS
In this report, a traffic sign recognition approach built on RECOGNITION USING A DYNAMIC PIXEL AGGREGATION
TECNHIQUE IN THE HSV COLOR SPACE”, 11th International
deep learning is presumed. This approach is efficient in Conference on Image Analysis and Processing, ISBN:0-7695-1183-X;
detecting as well as identifying traffic signs by combining DOI: 10.1109/ICIAP.2001.957071
picture pre-processing, detection of traffic signs, recognition [20] Kiran C. G., Lekhesh V. Prabhu, Abdu Rahiman V. and Rajeev K.,
along with classification. By utilising CNN, machine learning “Traffic Sign Detection and Pattern Recognition using Support Vector
Machine”, 2009 Seventh International Conference on Advances in
techniques, deep learning architecture, and several Python and Pattern Recognition, ISBN:978-0-7695-3520-3; DOI:
Tensor-Flow libraries, we are developing a model that is faster, 10.1109/ICAPR.2009.58
more accurate, requires less time, and is unaffected by varied [21] R. Nuthakki, P. Masanta and Y. T N, "Speech Enhancement based on
lighting and weather conditions. It may be put in any car with Deep Convolutional Neural Network," 2021 Fifth International
Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud)
the bare minimum of components. (I-SMAC), 2021, pp. 1-6, doi: 10.1109/I-SMAC52330.2021.9640736.
[22] Nuthakki, R., Masanta, P., Yukta, T.N. (2022). A Literature Survey on
Speech Enhancement Based on Deep Neural Network Technique. In:
REFERENCES Kumar, A., Mozar, S. (eds) ICCCE 2021. Lecture Notes in Electrical
Engineering, vol 828. Springer, Singapore. https://doi.org/10.1007/978-
[1] J.Stallkamp, M.Schlipsing, J.Salmen, C.Igel, “Man vs. computer:
Benchmarking machine learning algorithms for traffic sign 981-16-7985-8_2
[23] R. Nuthakki, A. S. Murthy and D. Naik, "Single channel speech
recognition”, 2012
[2] Manasa R., K. Karibasappa, Manoj Kumar Singh, Differential enhancement using a new binary mask in power spectral domain,"
Evolution Evolved RBFNN based automated recognition of Traffic 2018 Second International Conference on Electronics, Communication
and Aerospace Technology (ICECA), 2018, pp. 1361-1366, doi:
Sign Images, (2021)
[3] Danyah A. Alghmgham, Ghazanfar Latif, Jaafar Alghazo, Loay 10.1109/ICECA.2018.8474842.
Alzubaidi, Autonomous Traffic Sign (ATSR) Detection and [24] Nuthakki, R., Abbas, J., Afnan, A., Shariff, F.A., Hari, A. (2021).
Single-Channel Speech Enhancement Based on Signal-to-Residual
Recognition using Deep CNN, (2019)
[4] Rubén Laguna, Rubén Barrientos, L. Felipe Blázquez, Luis J. Miguel, Selection Criterion. In: Saini, H.S., Sayal, R., Govardhan, A., Buyya, R.
Traffic sign recognition application based on image processing (eds) Innovations in Computer Science and Engineering. Lecture Notes
in Networks and Systems, vol 171. Springer, Singapore.
techniques, (2014)
[5] Ying Sun, Pingshu Ge, Dequan Liu, “Traffic Sign Detection and https://doi.org/10.1007/978-981-33-4543-0_56
Recognition Based on Convolutional Neural Network”, 2019 Chinese
Automation Congress (CAC), ISBN:978-1-7281-4095-7; DOI:
10.1109/CAC48633.2019.8997240
[6] L. Priese, R. Lakmann, V. Rehrmann, “Ideogram identification in a
realtime traffic sign recognition system”, Proceedings of the Intelligent
Vehicles '95. Symposium, ISBN:0-7803-2983-X; DOI:
10.1109/IVS.1995.528299
[7] Yong-Jian Zheng, W. Ritter, R. Janssen, “An adaptive system for
traffic sign recognition”, Proceedings of the Intelligent Vehicles '94
Symposium, ISBN:0-7803-2135-9; DOI: 10.1109/IVS.1994.639496
[8] J. Miura, T. Kanda, Y. Shirai, “An active vision system for real-time
traffic sign recognition”, ITSC2000. 2000 IEEE Intelligent
Transportation Systems. Proceedings (Cat. No.00TH8493), ISBN:0-
7803-5971-2; DOI: 10.1109/ITSC.2000.881017
[9] Y.B. Damavandi, K. Mohammadi, "Speed limit traffic sign detection
and recognition", IEEE Conference on Cybernetics and Intelligent
Systems, Singapore, 2004, ISBN:0-7803-8643-4; DOI:
10.1109/ICCIS.2004.1460690

639
Authorized licensed use limited to: Consortium - Algeria (CERIST). Downloaded on November 20,2023 at 20:44:27 UTC from IEEE Xplore. Restrictions apply.