Proceedings of the 7th International Conference on Trends in Electronics and Informatics (ICOEI 2023)

IEEE Xplore Part Number: CFP23J32-ART; ISBN: 979-8-3503-9728-4

Traffic Sign Recognition System

for Autonomous Vehicles using Deep Learning

M s. S. JENCY, STUDENT/CSE M s. S. KARTHIKA, STUDENT/CSE M r. J. AJAYKUM AR, STUDENT/CSE

Dr.N.G.P iTech Dr.N.G.P iTech Dr.N.G.P iTech
2023 7th International Conference on Trends in Electronics and Informatics (ICOEI) | 979-8-3503-9728-4/23/$31.00 ©2023 IEEE | DOI: 10.1109/ICOEI56765.2023.10125896

Coimbatore, India Coimbatore, India Coimbatore, India

[email protected] [email protected] [email protected]

M r. R. SELVARAJ, AP/CSE, M s. A. P. AARTHI, STUDENT/CSE

Dr.N.G.P iTech Dr.N.G.P iTech
Coimbatore,India Coimbatore, India
[email protected] [email protected]

Abstract— The architecture of a fully autonomous car processed by utilizing several image processing techniques,
must be incorporated with a traffic sign recognition system. such as Contouring, Fit Ellipse, and Laplacian of
The Traffic Sign Recognition (TS R) consists of two Gaussian filtering. The following stage is to work with
components: detection and classification. The proposed study,
Neural Net works to remark the traffic sign arrangements.
which focuses on identifying these signals, is based on LIS A
dataset, which is the largest publicly accessible collection of The system is educated and assessed to identify the best
images of traffic signs in the world. The performance of both network design. The outcomes show the computational cost
aggregate channel features-based and integral channel feature- of the suggested technique as well as the incredibly precise
based detection approaches has reached its quality. In the classifications of traffic sign patterns with intricate backdrop
proposed study, the performance of Convolutional Neural images. Driving safety can only be ensured with the aid of a
Networks (CNN), aggregate channel characteristics, and traffic sign detection system. These solutions make it easier
integral channel features are evaluated. The proposed study for drivers to see the signs and follow them. In the
investigates the detection performance of CNN by tuning the recommended design, the dataset is initially trained using
convolutional layers, max-pool layers, and linear layers. The
CNN. An image which the user uploads serves as input to a
effectiveness of the proposed detection model is tested using the
trained model on which categorization must be done is
PAS CAL measure, a typical statistic for this system.
offered through a Graphical User Interface. CNN is
Keywords— Driverless vehicles, Indian traffic signs fundamentally limited by the need that each photo in a set of
classification, sign detection, image pre-processing, recognizing data have the same size before being fed. To do this, pre-
signs in smart vehicles, Traffic Sign Recognition, Region of processing is utilized to uniformly resize each image. To
Interest. categorize the photos in the collection, features from the
photographs are picked out and extracted.
The first usually involves applying image processing to
I. INTRODUCTION identify traffic signs in a series of videos or images. The
Traffic Sign Recognition (TSR) is a tool used for second one deals with the issue of efficiency in art ificial
controlling signs, alerting handlers, and directing or neural networks and is connected to the detection of these
forbidding unique behaviours. A rapid, dependable, discovered indications. The segmentation of shapes or
automatic traffic sign recognition and identification system colours is typically the basis for detection algorithms. The
can significantly increase the driver's convenience and collected segmented potential areas are fed into the
security. Additionally essential for autonomous driverless recognition step. In the system, the effectiveness and rapidity
vehicles or driver assistance systems is the automatic of the detection play a crucial part. Different techniques for
recognition of traffic signs. In order to detect traffic sign automatically identifying traffic signs have been developed
figures, this study uses neural networks. The images are pre- and have produced encouraging results. An exact
representation of the technology utilised in traffic symbol

979-8-3503-9728-4/23/$31.00 ©2023 IEEE 1116

Authorized licensed use limited to: BMS College of Engineering. Downloaded on April 05,2024 at 03:49:42 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the 7th International Conference on Trends in Electronics and Informatics (ICOEI 2023)
IEEE Xplore Part Number: CFP23J32-ART; ISBN: 979-8-3503-9728-4

identification is Neural Networks. The automated used to implement the neural network's training. The
recognition of objects is one area in particular where several proposed model's reliability is 93.58.
neural network applications have been created.
Anastasia Perkova, et. al [2], In this work, a method for
The effectiveness of the system in live time is affected by recognising traffic signs on video was built. This system uses
the challenges of traffic sign recognition and identification. a Convolutional Neural Network design known as YOLOv3.
The system must have highly effective algorithms and high- After 1000 iterations, the accuracy of identifying traffic signs
performance hardware. The environmental constraints also reached a maximu m of 93.91%, and after 2000 iterations, it
included illu mination, darkness opacity, air pollution, climate reached a maximu m of 99.94%.
(sunny, rainy, cloudy, etc.), and other visual distortions
Siwei Fan [3], This study suggests a detection and
including motion blur, vehicle shaking, and rapid contrast
identification system that uses an eight-layer convolutional
shifts that may occur regularly in the actual system.
neural network to train on various traffic signs to learn
various types of information. The training significantly
improved our detection and identification system, enabling it
II. TSR OVERVIEW to be more thorough and accurate than the majority of
Although the identification co mponent of TSR systems is current systems. This solved the issues of restricted accuracy
the focus of this study, a description of the entire system and performance when subjected to environmental impacts.
will be helpful. Typ ically, when writers talk about TSR
L.Mary Gladence, et. al [4], In order to construct
systems, they are talking about the detection of traffic signs,
Intelligent Vehicles IV, it is essential to forecast traffic
categorization of those signals after they have been found,
indicators. In order to develop intelligent vehicles for street
and tracking of those signs. The detection job solely
security, it is essential to identify and comprehend the street
considers determin ing the location of the traffic sign in an
indicators. The amount is based on CNN's earlier acquisition
image when provided an input of a photograph or a video
of LeNet-5. The method was taught using Gabor's
sequence. Classification, on the other hand, solely considers
presentation, followed by the customary transformation, and
what kind of road sign the observed traffic sign is. The third
then the parts were reconciled. The best tactic to apply in this
element is tracking, which keeps track of the indication that
situation is the Adams method. The value of shadow
has been detected fro m frame to frame. Different strategies
represents the traits of quick finding and low light. The street
can be used to address each component on its own. But for a
identifiable proof method is now being evaluated with the
whole TSR system to function, each co mponent will rely on
assistance of German Vehicle Identification. The suggested
the others. We will only examine the detection job in further
research project used HTML and CSS to design and create a
detail as it is the subject of our study. This shows a
user interface framework, and Flask was used to deploy the
straightforward representation of the Detection problem.
both model and the framework. The UI framework is meant
The detection process in a very rudimentary block diagram,
to accept the data needed to estimate the water quality, feed
however most methods will have some variation on this
it to a machine learning model, and then show the anticipated
theme. They generally consist of three stages: picture pre-
results on the page.
processing, feature extract ion, and detection. In order to get
ready for the feature ext raction process, the picture is Yi Chen, et. al [5], International Symposium on
cleaned up or transformed during the image pre-processing Co mmercial Electronics, 31st IEEE, 2022 (ISIE) Traffic sign
stage. Edges, Harr-like features, and colour features, for recognition, a part of the smart transportation system, has
example, are collected fro m the picture during the feature become increasingly popular among intelligent
extraction step and delivered to the detector. The detector is transportation researchers as a result of the advancement of
dependent on the methodology we choose. Typically, two autonomous driving technology. The majority of the TSR
approaches are used: a learning-based approach and a techniques now in use are built on deep learning, however
they are frequently constrained by the conflict between the
model-based approach. Both will be covered in greater
model's real-t ime performance and dependability, making it
detail. The detector's job is determining the location of the
challenging to use them extensively. Target identification
traffic sign in the image using the characteristics as input.
must meet rigorous accuracy and real-time standards in the
Although the detection stage is the main emphasis of this automated driving system. In particular, the image's modest
project, it is crucial to remember that a TSR system is fraction of traffic signs will make it harder to recognise them.
mean ingless without the categorization and tracking Traffic accidents can readily result fro m improper
components. information gathering and detection mistakes. Higher
standards for model size and computation are also necessary
to support rollout on in-vehicle devices. Therefore, creating a
III. LITERATURE SURVEY thin neural network with quick detection and high
Abhishek Kapoor, et. al [1], Convolutional Neural identification accuracy is crucial for the advancement of
Networks are already being used to handle an increasing autonomous driving. We conducted research on the road sign
number of object identification issues (CNN). The bulk of dataset known as TT100K and suggested an enhanced
new and existing computer vision occupations have YOLOv5 for road sign identification in this work. The
improved because to CNN's excellent performance and findings demonstrate that our technique performs
identification rate. In this research, a system for identifying exceptionally well in TSR and that overall accuracy has
traffic signs is developed using a Convolution Neural increased, particularly for the detection of small-sized
Network. The research also analyses and compares a variety objects. The map has increased by 6.68% since the upgrade.
of CNN arch itectures. Tensor flow and the Keras library are Additionally, the suggested network can fulfil the demands

979-8-3503-9728-4/23/$31.00 ©2023 IEEE 1117

Authorized licensed use limited to: BMS College of Engineering. Downloaded on April 05,2024 at 03:49:42 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the 7th International Conference on Trends in Electronics and Informatics (ICOEI 2023)
IEEE Xplore Part Number: CFP23J32-ART; ISBN: 979-8-3503-9728-4

for identification speed on equipment placed on vehicles and the classifier. By combining features with the CNN
can achieve recognition speeds of up to 91 frames per second classifier, the system is effectively generalised and made
(FPS), which has a significant practical value. resilient to changes in size, rotation, translation, and
Sri Suvetha C. S, et. al [6], The world is undergoing a illumination.
revolution in the field of artificial intelligence as more The imp lementation can be done by taking datasets of
technical applications integrate AI due to its increased Indian traffic signs fro m suitable sources. The images can be
effectiveness and efficiency. In several industries, including of any angles, colours, etc. The step-by-step process of the
manufacturing, robotics, transportation, healthcare, project is as follows:
agriculture, and the military, AI has become incredibly
pervasive and is now deeply influencing locker roo m Image acquisition: First, the image of the road scene is
choices. AI significantly enhances our lives, but it also captured using a camera or sensor.
frequently saves our lives. One of the most effective uses of
AI is in autonomous vehicles, or so-called self-driving Pre-processing: The acquired image is pre-processed to
automobiles, which play a crucial role in enabling the remove any noise or distortions that may affect the accuracy
machine to function independently by seeing and of the recognition process. This includes operations like
understanding the real-life surroundings. This article cropping, resizing, and colour normalization.
discusses the implementation of a voice-activated automatic
ROI extraction: Next, the Regions of Interest (ROI) are
traffic sign detecting system, one of the uses of autonomous
identified within the pre-processed image. These ROIs
cars that can considerably increase driving comfort and
correspond to areas of the image that are likely to contain
safety by distracting the driver fro m confusing traffic
situations. For better performance accuracy, the method and traffic signs. This can be done using techniques like object
database used must be suitable. Thus, in order to save detection or image segmentation.
computing resources, this research examines the Bounding box indication: Once the ROIs are identified, a
characteristics, efficacy, and efficiency of several algorithms bounding box is drawn around each ROI to indicate its
for deep learning and develops a variety of models. location within the image.
Riadh Ayachi, et. al [7], Another of the most crucial Feature extraction: The next step is to extract features
methods for providing support to people is the advanced
fro m each ROI. This involves identifying key characteristics
safety system (ADAS). By giv ing them important knowledge
of the traffic sign, such as its shape, colour, and texture.
about the environment's items, it helps the drivers handle the
car. In this study, we suggest an ADAS application for Feature extraction techniques can include Histogram of
recognising traffic signs. The deep learning method serves as Oriented Grad ients (HOG), Local Binary Patterns (LBP), or
the foundation for the suggested application. In specifically, Deep Convolutional Neural Networks (CNNs).
we processed the information fro m the system cameras using Classification: Once features have been extracted from
convolutional neural networks. The suggested CNN was each ROI, a classification algorith m is used to determine the
scaled to achieve a small model size without sacrificing class of traffic sign within the ROI. Here we use neural
accuracy. The suggested CNN maintains good speed and network classifier
real-time processing while being appropriate for embedded
implementation. When run on an Nvidia GTX 960 GPU, the Post-processing: Finally, the results of the classification
suggested CNN performs with an accuracy of 99.32% and an are post-processed to remove any false positives or
inference speed of 250 FPS on the European dataset. The misclassifications. This can be done using techniques like
outcomes demonstrated the effectiveness of the scaling non-maximum suppression or thresholding.
method. To acquire a compact model size and great
performance, this method is excellent.
All the above works were analysed and the disadvantages
of them are mentioned as below:
• High training time is required by the models.
• Low accuracy obtained from the evaluations.
• Low stability on the usage of the models.
• Huge number of datasets is required by the
models.

Figure No.4.1. Block Diagram

IV. METHODOLOGY
An effective method for classifying Indian traffic signs
was suggested in this effort. We build a classification system
A. PRE - PROCESSING:
for Indian traffic signs based on the data available. The
classification of traffic indicators in this work also considers Images are made up of many sizes and shapes. Also, they
recognition and identification. To co mpletely fulfil the aim, have many sources. Any pictorial data must be pre-
the features are gathered from the data on Indian traffic processed in order to extract all these variations into
signs, and the CNN algorith m is then instructed to construct consideration. The most frequently used format, RGB, used

979-8-3503-9728-4/23/$31.00 ©2023 IEEE 1118

Authorized licensed use limited to: BMS College of Engineering. Downloaded on April 05,2024 at 03:49:42 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the 7th International Conference on Trends in Electronics and Informatics (ICOEI 2023)
IEEE Xplore Part Number: CFP23J32-ART; ISBN: 979-8-3503-9728-4

to render the most " natural pictures". The initial method in In other words, you may select a certain reg ion of interest in
pre-processing of data is generating the images analogous a picture and do certain activit ies there by utilizing this
size. In this case, we utilized bus resizing to change the function.
images in the data- set to an analogous resolution.
D. CONTOUR:
B. FEATURE EXTRACTION: A contour is a line that joins all the continuous points (along
The feature extraction approach is beneficial when less the border) and has the same color or brightness. In the
processing power is needed without losing important or second, we search for the boundary in a binary image after
relevant data. Additionally, by min imizing the amount of finding the counter in the binary picture. The binary images
extraneous data, feature extraction might aid an analysis. emp loy threshold or clever edge detection to ensure
The research and understanding phases of the machine correctness. Finding the contour in a b inary image using
learning process are also accelerated by the system's OpenCV is analogous to separating a white object fro m a
attempts to develop various features and the reduced data. black background. The contour in the b inary image can be
located using the findContours() function implemented by
OpenCV.
E. CNN:
To ext ract informat ion fro m a p iece of data, convolutional
neural networks (CNNs) interact with the data. Like photos,
sounds, or v ideos, CNN can use them for data ext raction. In
CNN, three items domineer. The first is local receptive field,
which is then trailed by shared weight and biases, and then
by activation and pooling. Neural networ ks must in itially be
trained on a vast quantity of data to extract the feature of an
input for CNN. Prior to being shown as the result, an input
goes through picture pre-processing, ext raction of features
by a predefined set, classification of data, and th en the
outcome. The CNN can only process inputs for which
the neural network has been tutored and data has been
saved. They are used in image classification, video detection
and recognition. The accuracy is calculated using the
following formulae:

Definiteness= TP/(TP+FP) (1)

Acuteness=TP/(TP+FN) (2)
Distinction=TN/(TN+FP) (3)
Veracity=(TP+TN)/(TP+TN+FP+FN) (4)

Where TP= True Positive,

FP= False Positive,
TN= True Positive,
Figure No.4.2. Data Flow Diagram
FN= False Positive

C. OPENCV ROI: V. CONCLUSION:

To accurately split and ext ract the traffic sign ROI is used.
This proposed system describes a TSR system that can
We may manually and precisely select the region of interest
classify the sign from a specific picture taken by a car
we need fro m the image and perform a range of actions for
camera. The three actions of image analysis are detecting,
that specific area using the OpenCV feature. That exact site
segmenting, and classifying. Segmenting separates the sign
could be a source of information for another work. Using
fro m backdrop after detecting the edges of the sign. This
the coordinates, we can also create a tracking shape on the
model offers a simple feature selection method using data
area or accurately and freely crop an image. First, we must
sets utilized for the recognition and identification of traffic
import the necessary libraries, in this case OpenCV and
signs. These datasets provide multip les of complex traffic
Nu mPy. This applicat ion relies heavily on the Nu mPy
signs, including scenic maps and tilted images, asymmetrical
module because OpenCV uses it as the foundation for all
illu mination, d istracted, and disrupted signs. Many
image man ipulation operations. It should go without saying
complicated and individualized traffic signs are used to
that reading an image before doing any of its nu merous
check the working of the algorith m. The working of the
operations and saving it in a variab le so you may use it later
system is tested against a range of sophisticated and
for revisions to choose the ROI function's actual function().

979-8-3503-9728-4/23/$31.00 ©2023 IEEE 1119

Authorized licensed use limited to: BMS College of Engineering. Downloaded on April 05,2024 at 03:49:42 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the 7th International Conference on Trends in Electronics and Informatics (ICOEI 2023)
IEEE Xplore Part Number: CFP23J32-ART; ISBN: 979-8-3503-9728-4

complicated traffic signs using Convolution Neural Network. [6].Everingham, M., Gool, L., Williams, C., Winn, J., and Zisserman,
A. (2010). The Pascal Visual Object Classes (VOC) Challenge. Int. J.
Traffic sign recognition is a critical co mponent of advanced
Comput. Vision 88, 2 (June 2010), 303-338.
driver assistance systems and autonomous vehicles. It DOI=http://dx.doi.org/10.1007/s11263-009-0275-4
involves the automatic detection and classification of traffic
signs using computer vision techniques. Over the past few [7].Freund, Y. and Schapire, R.E. (1996), Experiments with an new
years, significant progress has been made in this field due to boosting algorithm, Machine Learning: Proceedings of the Thirteenth
International Conference, Morgan Kauffman, SanFrancisco, pp.148 -
advancements in deep learning and convolutional neural 156 (1996). Introduced AdaBoost
networks (CNNs). However, there are still several challenges
that need to be addressed, such as the detection of occluded [8].Friedman, J., Hastie, T., and Tibshirani, R. (2000), “ Additive
or damaged signs, recognition of signs under different LogisticRegression: A Statistical View of Boosting,” Annals of
lighting conditions, and the robustness of the algorithms Statistics,vol. 38, no. 2, pp. 337-374, 2000.
against environmental factors.
[9].Forsyth, D. and Ponce, J. (2021). Computer Vision a Modern
Approach (2nd ed., p. 549). New York, New York, USA: Pearson

VI. FUTURE WORK: [10].Houben, S., Stallkamp, J., Salmen, J., Schlipsing, M., and Igel,
C. (2013), "Detection of traffic signs in real-world images: The
To improve the accuracy and robustness of TSR systems, German traffic sign detection benchmark," Neural Networks
future research can focus on the following areas: (IJCNN), The 2013 International Joint Conference on, Dallas, TX,
2013, pp. 1-8. doi: 10.1109/IJCNN.2013.6706807
Multi-task learning: Incorporating mult iple tasks such as
object detection, segmentation, and classification to a
[11].Krizhevsky, A., Sutskever, I., and Hinton, G. (2021), Imagenet
network can improve the overall performance of the system. classification with deep convolutional neural networks. In NIPS,
2021.
Domain adaptation: Developing algorithms that can adapt
to different environments, such as different lighting
[12].Li, F., Karpathy, A., and Johnson, J. (2015), "CS231n:
conditions or weather conditions, can enhance the system's Convolutional Neural Networks for Visual Recognition," www.
robustness. http://cs231n.stanford.edu/
Fusion of mult iple sensors: Integrating information fro m
[13].Liang, M., Yuan, M., Hu, X., Li, J., and Liu, H. (2013) “Traffic
mu ltiple sensors, such as cameras and lidars, can provide sign detection by ROI extraction and histogram features-based
more robust and accurate results. recognition,” in Proc. IEEE IJCNN, Aug. 2013, pp. 1–8.
Exp lainability: Developing algorithms that can provide
[14].Lim, K., Lee, T., Shin, C., Chung, S., Choi, Y., and Byun, H.,
exp lanations for the decisions made by the system can (2014). Real-time illumination-invariant speed-limit sign recognition
increase trust in the technology and facilitate its adoption. based on a modified census transform and support vector machines.
In Proceedings of the 8th International Conference on Ubiquitous
Real-time performance: Designing algorithms that can Information Management and Communication (ICUIMC '1 4). ACM,
achieve real-time performance on low-power hardware can New York, NY, USA, , Article 92 ,5 pages.
make TSR systems more practical and cost-effective. DOI=http://dx.doi.org/10.1145/2557977.2558090

In conclusion, TSR is an important research area with [15].Liu, W., Wu, Y., Lv, J., Yuan, H., and Zhao, H. (2021) "U.S.
significant potential for improving road safety and advancing speed limit sign detection and recognition from image sequences,"
autonomous vehicle technology. Continued research and Control Automation Robotics & Vision (ICARCV), 2021 12th
development in this field will be critical for International Conference on, Guangzhou, 2021, pp. 1437-1442.
doi: 10.1109/ICARCV.2021.6485388
achieving these goals.
[16].Loy, G. and Barnes, N. (2004), “Fast shape-based road sign
detection for a driver assistance system,” in Proc. IEEE/RSJ Int. Conf.
VII. REFERENCES IROS, 2004, vol. 1, pp. 70–75
[1].Bishop, C. (2007). "Pattern Recognition and Machine Learning"
(2nd ed). New York, New York, USA: Springer. [17].Mogelmose, A., Liu, D., and Trivedi, M. M. (2015), "Detection
of U.S. T raffic Signs," in IEEE Transactions on Intelligent
Transportation Systems, vol. 16, no. 6, pp. 3116-3125, Dec. 2015.
[2].Dalal, N. and Triggs, B. (2005) "Histograms of oriented gradients doi: 10.1109/TITS.2015.2433019
for human detection," Computer Vision and Pattern Recognition,
2005. CVPR 2005. IEEE Computer Society Conference on, San [18].Mogelmose, A., Liu, D., and Trivedi, M. M. (2014), "T raffic sign
Diego, CA, USA, 2005, pp. 886-893 vol. 1. doi: detection for U.S. roads: Remaining challenges and a case for
10.1109/CVPR.2005.177 tracking," Intelligent Transportation Systems (IT SC), 2014 IEEE 17th
International Conference on, Qingdao, 2014, pp. 1394-1399. doi:
[3].Dollár, P., “Piotr’s Computer Vision Matlab Toolbox (PMT).” 10.1109/ITSC.2014.6957882
[Online]. Available:
http://vision.ucsd.edu/\ignorespacespdollar/toolbox/doc/index. html [19].Mogelmose, A., Trivedi, M. M., and Moeslund, T. B. (2021),
"Vision-Based Traffic Sign Detection and Analysis for Intelligent
[4].Dollar, P., Appel, R., Belongie, S., and Perona, P. (2014), "Fast Driver Assistance Systems: Perspectives and Survey," in IEEE
Feature Pyramids for Object Detection," in IEEE Transactions on Transactions on Intelligent Transportation Systems, vol. 13, no. 4, pp.
Pattern Analysis and Machine Intelligence, vol. 36, no. 8, pp. 1532- 1484-1497, Dec. 2021.
1545, Aug. 2014. doi: 10.1109/TPAMI.2014.2300479 doi: 10.1109/TITS.2021.2209421

[5].Dollár, P., T u, Z., Perona, P., and Belongie S. (2009), “Integral [20].Staudenmaier, A., Klauck, U., Kreßel, U., Lindner, F., and
channel features,” in Proc. BMVC, vol. 2, no. 3, p. 5, 2009 Wöhler, C. (2021) “Confidence measurements for adaptive Bayes
decision classifier cascades and their application to US speed limit

979-8-3503-9728-4/23/$31.00 ©2023 IEEE 1120

Authorized licensed use limited to: BMS College of Engineering. Downloaded on April 05,2024 at 03:49:42 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the 7th International Conference on Trends in Electronics and Informatics (ICOEI 2023)
IEEE Xplore Part Number: CFP23J32-ART; ISBN: 979-8-3503-9728-4

detection,” in Proc. Pattern Recognit., ser. Lecture Notes in Computer 10.1109/CVPR.2001.990517

Science, vol. 7476, pp. 478–487, 2021.
[23].Wang, G., Ren, G., Wu, Z., Zhao, Y., and Jiang, L. (2013), “A
[21].Vázquez-Reina, A., Lafuente-Arroyo, S., Siegmann, P., robust, coarse-to- fine traffic sign detection method,” in Proc. IEEE
Maldonado-Bascón, S., and Acevedo-Rodríguez, F. (2005), “Traffic IJCNN, Aug. 2013, pp. 1–5.
sign shape classification based on correlation techniques,” in Proc.
5th WSEAS Int. Conf. Signal Process., Comput. Geometry Artif. [24].United Nations (1978), "Vienna Convention on road signs and
Vis., 2005, pp. 149–154. signals,"

[22].Viola, P. and Jones, M. (2001), "Rapid object detection using a [25].Zimmerman, A (2021), "Category-level Localization", Visual
boosted cascade of simple features," Computer Vision and Pattern Geometry Group University of Oxford,
Recognition, 2001. CVPR 2001. Proceedings of the 2001 IEEE http://www.robots.ox.ac.uk/~vgg
Computer Society Conference on, 2001, pp. I-511-I-518 vol.1. doi:

979-8-3503-9728-4/23/$31.00 ©2023 IEEE 1121

Authorized licensed use limited to: BMS College of Engineering. Downloaded on April 05,2024 at 03:49:42 UTC from IEEE Xplore. Restrictions apply.