Volume-3, Issue-4, April 2025 International Journal of Modern Science and Research Technology

ISSN No- 2584-2706

Hardware Model-Based Traffic Sign Detection and

Recognition
Neeraj Kumar Das
Shubham Shukla
Shiva Gaur
Khushboo

Abstract
In the modern world, correctly recognizing and B+, Embedded System, Convolution Neural
categorizing traffic signs is essential for Network (CNN), Image Processing, Real-time
improving road safety and enabling autonomous Classification, Smart Vehicle System, Traffic
driving technologies. This study uses the Monitoring.
Raspberry Pi 3 Model B+ as the central
processing unit to provide a hardware-based, I. Introduction
real-time method for traffic sign detection and These days, advanced driver assistance systems
recognition. (ADAS) are an essential component of modern
Our system integrates image collection, cars, especially when it comes to identifying
preprocessing, and classification into a small, and comprehending traffic signs, which is a
embedded device intended for realistic, real- crucial area of visual computing [1]. Road signs
world deployment, in contrast to conventional provide vital information like speed limits,
software-heavy approaches. The system makes directions, and cautions that assist drivers in
use of a lightweight Convolutional Neural making safe and educated decisions. They act as
Network (CNN) to effectively categorize traffic a kind of global language on the road. Drivers
signs, employs a camera module to gather visual can improve their navigation skills, steer clear
data, and applies image enhancement techniques of possible dangers, and help create a safer and
to improve clarity. The CNN is perfect for low- more efficient traffic flow by correctly
power devices like the Raspberry Pi since it is interpreting these signs [2].
tuned to balance speed and accuracy. The effectiveness of traffic signs is greatly
Our test demonstrates that the system can influenced by their visual design.
accurately identify and detect a variety of traffic Characteristics like shape and color aren’t just
signs in a range of environmental and lighting aesthetic—they’re critical for fast recognition.
situations. All things considered, this study Most countries follow standardized patterns
provides a scalable, economical, and energy- using specific colors (like red, blue, and yellow)
efficient solution for intelligent automobiles and and shapes (such as circles, triangles, and
smart traffic systems. In order to achieve even rectangles) to ensure uniformity and quick
greater performance, future advancements will understanding across different regions. These
concentrate on increasing recognition accuracy visual cues help both humans and machines
using more sophisticated deep learning models identify signs quickly, even at a glance.
and utilizing cutting-edge AI technology. However, conditions on the road are rarely
perfect. Factors such as poor weather, low light,
Keywords or worn-out signage can make it difficult to see
Traffic Sign Detection & Recognition, and interpret road signs clearly. These
Autonomous Navigation, Raspberry Pi 3 Model challenges highlight the need for reliable and

IJMSRT25APR044 www.ijmsrt.com 192

DOI: https://doi.org/10.5281/zenodo.15300745
Volume-3, Issue-4, April 2025 International Journal of Modern Science and Research Technology
ISSN No- 2584-2706

intelligent detection systems that can operate

accurately in real-world conditions [3, 4].
To tackle this, researchers have been developing
various advanced techniques to improve how
ADAS systems detect and understand traffic
signs [5]. Some of the most promising methods
involve deep learning, where models are trained
to recognize signs using existing datasets, even
if those datasets are relatively small. Others use
approaches like multi-scale feature fusion,
which enhances the system’s ability to detect
how large or little they appear in a picture [6].
To put it simply, the continuous research and
development of traffic sign analysis is not
merely a technological undertaking; rather, it is Fig. 1. Block diagram of Traffic Sign
an essential step in creating safer roads for all. Detection and Recognition
ADAS helps drivers stay vigilant, make smarter
judgments, and ultimately lower the risk of II. Related Work
accidents on the road by fusing smart Many researchers have looked into different
technology with practical applications. ways to improve how traffic signs are detected
The process of identifying road signs is often and recognized, aiming to make the process
divided into two primary steps in the majority of more accurate and efficient. One interesting
current systems: determining the type of sign method combines detection, tracking, and
and estimating its potential location. This recognition using techniques like Adaboost,
enhances the system's accuracy and color-sensitive Haar wavelet features, and the
effectiveness. use of motion (temporal) data to figure out
In this study, a new and more effective speed limits [7]. This system was trained on a
technique has been presented to increase the dataset with around 4,000 images, covering 23
speed and precision of traffic sign detection. types of traffic signs. The number of examples
The system goes through a few image for each sign type ranged from 30 to 600. To
preprocessing processes before it even starts to check how well the system worked, a separate
recognize signs. These procedures aid in image set of 1,700 images was used for testing—and it
preparation and cleaning, which facilitates the reached an impressive accuracy of 94%.
system's ability to comprehend and accurately Another study [8] focused on using a neural
identify the indicators. network approach to recognize single-digit
A detailed workflow has been developed to speed limit signs found on roads in Europe and
clarify the entire procedure. This well-organized the U.S. This method was built using the MAPS
graphic demonstrates how every component of software framework. However, it didn't provide
the system cooperates to achieve the end result. specific classification results for each type of
This figure, which is displayed in Figure 1, sign. The system, which included both tracking
provides a visual representation of the entire and detection features, was tested on 281
procedure, from data preparation to precise different traffic signs.
traffic sign detection and identification. In [9], a traffic sign classification model based
on color segmentation and shape-based
verification was developed. The detection
process relied on selecting an appropriate neural

IJMSRT25APR044 www.ijmsrt.com 193

DOI: https://doi.org/10.5281/zenodo.15300745
Volume-3, Issue-4, April 2025 International Journal of Modern Science and Research Technology
ISSN No- 2584-2706

network according to the sign’s shape and RGB techniques, this integrated strategy performed
color properties. better, making it more practical.
A different study [10] employed a multi-layer Processed images were stored in the Portable
perceptron neural network, which was trained Pixel Map (PPM) format without compression.
on a dataset containing 2,880 images. This Each image was carefully annotated to indicate
model was further tested on 1,233 images of the presence and type of visible traffic signs. All
speed limit signs, achieving an accuracy rate of label information, including class identifiers and
92.4%. However, the research did not specify bounding box data, is compiled in a structured
whether the same traffic sign instances were CSV file to support supervised learning models.
repeated across different image groups.
Another method [11] utilized a dataset B. Dataset Characteristics and Relevance to
containing 1,300 pre-processed images from six the Proposed System
categories five corresponding to speed limits In the present work, the GTSRB dataset [20] is
and one representing noise. The study tested adopted as a benchmark to train and validate the
various segmentation-based techniques using traffic sign classification model due to its
binary image processing. Additionally, research comprehensive structure and real-world
in [12] compiled an extensive dataset featuring diversity. The dataset comprises over 50,000
36,000 images of Spanish traffic signs from 193 annotated images categorized into 43 distinct
classifications, leveraging pictogram traffic sign classes. It presents a wide range of
identification techniques. A different approach visual variations, including differences in
[13] explored the use of histogram features and orientation, illumination levels, weather
Region of Interest (ROI) extraction for traffic conditions, partial occlusions, and spatial
sign recognition. However, the results obtained positioning of signs within each frame—factors
in this study were below expectations, with the that closely resemble real-time driving
proposed methodology failing to achieve the environments.
desired performance levels. The complete dataset includes 51,840 images,
A direct comparison of the studies mentioned with dimensions ranging from 15×15 to
above is difficult, as many of them utilized 250×250 pixels. The traffic signs within these
proprietary datasets, restricting accessibility for images are not consistently centered, which
validation by other researchers. To address this introduces natural complexity for detection
challenge, a study in [14] applied the publicly tasks. The dataset is systematically divided into
available GTSRB dataset to develop a novel two parts: 39,209 images are reserved for
Convolutional Neural Network (CNN)-based training, while the remaining 12,630 images
architecture for traffic sign detection and form the testing set. To prevent data leakage,
classification. Faster R-CNN for multi-object images containing the same physical sign are
recognition was used in another study [15], confined to a single subset. Additionally, a
which included various datasets with small portion of the training data is allocated for
automobiles, traffic signs, and traffic lights. model validation and performance tuning.
CNNs were used by the authors in [16] to create Each image is linked to a specific traffic sign
a technique specifically intended for identifying label, such as "Speed Limit 50 km/h," "Speed
Indian traffic signs. Building on this, a real-time Limit 60 km/h," or "Road Work." The traffic
detection system that can detect both audio and signs are divided into four primary categories—
video was developed in another study [17]. It Warning, Prohibitory, Mandatory, and
analyzes auditory cues using the VGG model Others—in order to better arrange them and
and uses YOLOv5 to identify indications in make them easier to use with various
video data. When compared to earlier recognition algorithms. Sample photos from

IJMSRT25APR044 www.ijmsrt.com 194

DOI: https://doi.org/10.5281/zenodo.15300745
Volume-3, Issue-4, April 2025 International Journal of Modern Science and Research Technology
ISSN No- 2584-2706

each of these categories are displayed in Figure

2. The hardware arrangement in this study, Fig. 2. Representative Classes of Traffic
which consists of the OV5647 5MP 1080P IR- Signs Included in the GTSRB Dataset
CUT camera module coupled to a Raspberry
Pi 5 Model, is also tested using these samples IV. Embedded Hardware Integration and
as benchmarks. Software Stack
A. Raspberry Pi 5 Model
III. Dataset In order to better handle demanding tasks like
Images from a number of visits taken in the real-time picture processing and traffic sign
spring and fall of 2010—mostly in the Bochum recognition, the Raspberry Pi 3 Model B was
region of Germany—are included in the utilized in this study with an improved 4 GB
GTSRB collection. The data was collected in a RAM arrangement. With a 64-bit quad-core
variety of illumination circumstances, including ARM Cortex-A53 processor clocked at 1.4
brilliant daylight and low-light dusk, and covers GHz, this little computer offers a solid mix
a broad range of driving locations, such as city between performance and energy economy,
streets, country roads, and highways. The making it ideal for computer vision and artificial
dataset is extensive and varied for training and intelligence projects.
testing traffic sign recognition algorithms The board can function both locally and
because it also includes a variety of weather remotely thanks to its integrated support for
situations. Figure 2 displays a few exemplary Ethernet, Bluetooth 4.2, and Wi-Fi (802.11
samples. The traffic signs included in the b/g/n/ac). Additionally, it has a full-size HDMI
dataset follow the guidelines established by the and four USB 2.0 output connections, and a 40-
Vienna Convention on Road Signs and pin GPIO header for attaching other
Signals, ensuring consistency and peripherals. One of the main features of this
standardization in their appearance. setup is its interoperability with the OV5647
5MP IR-CUT camera module, which is
A. Image Acquisition and File Structure connected directly to the Raspberry Pi via the
The GTSRB dataset was created using images dedicated CSI (Camera Serial Interface) port.
captured with a high-resolution OV5647 5MP To get the system operational, the operating
1080P IR-CUT camera, which was set to system is first installed onto a microSD card,
automatic exposure to adjust for different which is then inserted into the Raspberry Pi.
lighting conditions. These images were The setup process is completed by configuring
originally captured in Bayer pattern format. To wireless connectivity, setting up user
reconstruct accurate color images, an edge- credentials, and enabling remote access (such as
adaptive and constant-hue demosaicking SSH or VNC).
technique was applied, transforming them into To install the operating system, the official
RGB. Raspberry Pi Imager tool is used to flash the
OS image onto the microSD card. For this
project, a 32GB SanDisk microSD card was
used to ensure sufficient space for the operating
system, project files, libraries, and temporary
image storage used during traffic sign detection
and recognition tasks. The system's software
backbone is made up of crucial software
libraries like Open CV and Python, which are

IJMSRT25APR044 www.ijmsrt.com 195

DOI: https://doi.org/10.5281/zenodo.15300745
Volume-3, Issue-4, April 2025 International Journal of Modern Science and Research Technology
ISSN No- 2584-2706

compatible with the OS thanks to its Linux The OS is installed by flashing the image file
foundation. onto a microSD card using the official
It has a Camera Serial Interface (CSI) connection. Raspberry Pi Imager tool.
Clear images may be captured with this camera For this project, a 32GB SanDisk microSD
even in a variety of lighting situations. The Raspbian card was chosen to provide enough storage for
OS (based on Debian) that the Raspberry Pi runs on the operating system, necessary dependencies,
offers a reliable environment for executing Python
scripts and necessary libraries like OpenCV and
TensorFlow Lite, which are both utilized in this
study to identify and categorize traffic signs.

and the images captured during the project.

Fig. 5. Basic hardware features and

specification summary of the Raspberry Pi 5
Model.

IV. Result and Analysis

A Raspberry Pi 5 and a webcam were used to
effectively identify and recognize traffic signs
in real time. To identify and categorize traffic
signs, the trained model continually analyzes
camera picture frames. When no sign is
detected, the system displays the message "No
Sign Detected" on the output screen, as
illustrated in Fig. 4.
Testing and validation accuracies were noted at
Fig. 4. Selection window displaying available several training phases in order to assess model
Raspbian OS performance. As indicated in Table I, results
The system in this study runs on the 64-bit were taken every 50 epochs, and the highest
version of the Raspbian operating system
(Debian Bullseye), which is specially designed
for Raspberry Pi hardware. This version was
selected because it offers better performance
and works well with modern computer vision
libraries, making it ideal for real-time
processing. While Raspbian is available in both
32-bit and 64-bit versions, the 64-bit edition is performance was observed after 200 epochs.
more efficient at managing memory, which is
particularly beneficial for the 4GB RAM TABLE I. Accuracy for Testing and Validation
configuration of the Raspberry Pi 5 Model at Different Epochs
used in this research.  No Sign Detected

IJMSRT25APR044 www.ijmsrt.com 196

DOI: https://doi.org/10.5281/zenodo.15300745
Volume-3, Issue-4, April 2025 International Journal of Modern Science and Research Technology
ISSN No- 2584-2706

 Turn Right Sign

Fig. 13 (A). Output image showing right

(59.2%)

Fig. 10 (A). Output image showing

No Sign Detected

Fig. 13 (B). Output image showing

Detected: Right

Fig. 10 (B). Output image showing No Sign

Detected

In addition to the above, the system also

successfully detects key traffic signs used for
basic road navigation, including:
 Stop Sign

Fig.14. Complete hardware model of traffic sign

detection using Raspberry Pi

 Turn Left Sign

Fig. 11 (A). Output image showing

Stop (53.9%)

Fig. 11 (B). Output image showing Fig. 12 (A). Output image showing
Detected: Stop Left (59.28%)
Detected: Left

IJMSRT25APR044 www.ijmsrt.com 197

DOI: https://doi.org/10.5281/zenodo.15300745
Volume-3, Issue-4, April 2025 International Journal of Modern Science and Research Technology
ISSN No- 2584-2706

network," 2016 International Conference on

Fig. 12(B). Output image showing Intelligent Control Power and Instrumentation
(ICICPI), Kolkata, India, 2016, pp. 307-311.
Conclusion and Future Scope [3] Tao Chen and Shijian Lu, "Accurate and
In this research, a hardware-based system for Efficient Traffic Sign Detection Using
traffic sign detection and recognition was Discriminative AdaBoost and Support Vector
successfully developed and implemented using Regression," in IEEE Transactions on Vehicular
a Raspberry Pi 5 and the OV5647 5MP IR-CUT Technology, June 2016, vol. 65, no. 6, pp. 4006-
camera module. The system effectively 4015.
identifies and classifies various traffic signs in [4] Ma Xing, Mu Chunyang, Wang Yan, Wang
real-time, demonstrating its potential for Xiaolong and Chen Xuetao, "Traffic sign
integration into intelligent transportation detection and recognition using color
systems and advanced driver-assistance systems standardization and Zernike moments," 2016
(ADAS). The model provides dependable Chinese Control and Decision Conference
performance with minimal latency by utilizing (CCDC), Yinchuan, China, 2016, pp. 5195-
machine learning approaches and efficient 5198.
image processing algorithms, which makes it [5] Lu Wei, Lu Runge, Liu Xiaolei, "Traffic
appropriate for practical applications. sign detection and recognition via transfer
The implementation of a small, affordable learning," 2018 Chinese Control and Decision
hardware configuration highlights the viability Conference (CCDC), Shenyang, China, 2018,
of implementing such systems in embedded pp. 5884-5887.
settings, particularly when resources are limited. [6] Canyong Wang, "Research and Application
This work provides important insights into real- of Traffic Sign Detection and Recognition
time difficulties such as changing lighting Based on Deep Learning," 2018 International
conditions, motion blur, and hardware Conference on Robots & Intelligent Systems
restrictions, bridging the gap between software (ICRIS), Changsha, China, 2018, pp. 150-152.
simulations and actual implementation. [7] Md. Abdul Alim Sheikh, Alok Kole and
All things considered, the research demonstrates Tanmoy Maity, "Traffic sign detection and
the feasibility of fusing open-source software classification using color feature and neural
with reasonably priced hardware to provide network," 2016 International Conference on
responsive, intelligent systems for the detection Intelligent Control Power and Instrumentation
and recognition of traffic signs. Future (ICICPI), Kolkata, India, 2016, pp. 307-311.
improvements might incorporate deep learning [8] Tao Chen and Shijian Lu, "Accurate and
models for improved accuracy and adaptability, Efficient Traffic Sign Detection Using
grow the sign collection for wider use, and Discriminative AdaBoost and Support Vector
integrate GPS for location-based context. Regression," in IEEE Transactions on Vehicular
Technology, June 2016, vol. 65, no. 6, pp. 4006-
References 4015.
[1] Canyong Wang, "Research and Application [9] Ma Xing, Mu Chunyang, Wang Yan, Wang
of Traffic Sign Detection and Recognition Xiaolong and Chen Xuetao, "Traffic sign
Based on Deep Learning," 2018 International detection and recognition using color
Conference on Robots & Intelligent Systems standardization and Zernike moments," 2016
(ICRIS), Changsha, China, 2018, pp. 150-152. Chinese Control and Decision Conference
[2] Md. Abdul Alim Sheikh, Alok Kole and (CCDC), Yinchuan, China, 2016, pp. 5195-
Tanmoy Maity, "Traffic sign detection and 5198.
classification using color feature and neural

IJMSRT25APR044 www.ijmsrt.com 198

DOI: https://doi.org/10.5281/zenodo.15300745
Volume-3, Issue-4, April 2025 International Journal of Modern Science and Research Technology
ISSN No- 2584-2706

[10] Lu Wei, Lu Runge, Liu Xiaolei, "Traffic on pictogram identification for the road-sign
sign detection and recognition via transfer recognition task using SVMs,” ELSEVIER
learning," 2018 Chinese Control and Decision Computer Vision and Image Understanding,
Conference (CCDC), Shenyang, China, 2018, March 2010, Volume 114, Issue 3, pp. 373–383.
pp. 5884-5887. [18] Ming Liang, Mingyi Yuan, Xiaolin Hu,
[11] Yuan Yuan, Zhitong Xiong, Qi Wang, Jianmin Li and Huaping Liu, "Traffic sign
"VSSA-NET: Vertical Spatial Sequence detection by ROI extraction and histogram
Attention Network for Traffic Sign Detection," features-based recognition," The 2013
in IEEE Transactions on Image Processing, July International Joint Conference on Neural
2019, vol. 28, no. 7, pp. 3423-3434. Networks (IJCNN), Dallas, TX, USA, 2013, pp.
[12] C. Bahlmann, Y. Zhu, Visvanathan 1-8.
Ramesh, M. Pellkofer and T. Koehler, "A [19] R. Karthika, Latha Parameswaran, “A
system for traffic sign detection, tracking, and novel convolutional neural network-based
recognition using color, shape, and motion architecture for object detection and recognition
information," IEEE Proceedings. Intelligent with an application to traffic sign recognition
Vehicles Symposium, 2005., Las Vegas, NV, from road scenes”, Pattern Recognition and
USA, 2005, pp. 255-260. Image Analysis, July 2022, Volume 32, pp. 351-
[13] Fabien Moutarde, Alexandre Bargeton, 362.
Anne Herbin and Lowik Chanussot, "Robust on- [20] Kaushek Kumar T R, S Thiruvikkraman,
vehicle real-time visual detection of American Gokul R, Nirmal A and Karthika R, "Evaluating
and European speed limit signs, with a modular the Scalability of a Multi-Object Detector
Traffic Signs Recognition system," 2007 IEEE Trained with Multiple Datasets," 2021 5th
Intelligent Vehicles Symposium, Istanbul, International Conference on Intelligent
Turkey, 2007, pp. 1122-1126. Computing and Control Systems (ICICCS),
[14] Alberto Broggi, Pietro Cerri, Paolo Medici, Madurai, India, 2021, pp. 1359-1366.
Pier Paolo Porta, and Guido Ghisio, "Real Time [21] Rajesh Kannan Megalingam, Kondareddy
Road Signs Recognition," 2007 IEEE Intelligent Thanigundala, Sreevatsava Reddy Musani,
Vehicles Symposium, Istanbul, Turkey, 2007, Hemanth Nidamanuru and Lokesh Gadde,
pp. 981-986. "Indian traffic sign detection and recognition
[15] Christoph Gustav Keller, Christoph using deep learning", International Journal of
Sprunk, Claus Bahlmann, Jan Giebel and Transportation Science and Technology,
Gregory Baratoff, "Real-time recognition of Volume 12, Issue 3, 2023, pp. 683-699.
U.S. speed signs," 2008 IEEE Intelligent [22] V Sowbaranic Raj, Jalakam Venu Madhava
Vehicles Symposium, Eindhoven, Netherlands, Sai, N A Lakkshmi Yogesh, SB Kavya Preetha
2008, pp. 518-523. and Lavanya R, "Smart Traffic Control for
[16] Azam Sheikh Muhammad, Niklas Emergency Vehicles Prioritization using Video
Lavesson, Paul Davidsson, Mikael Nilsson, and Audio Processing," 2022 6th International
“Analysis of speed sign classification Conference on Intelligent Computing and
algorithms using shape-based segmentation of Control Systems (ICICCS), Madurai, India,
binary images”, International Conference on 2022.
Computer Analysis of Image and Patterns
(CAIP), Lecture Notes in Computer Science
(LNIP), Volume 5702, pp. 1220-1227, 2009.
[17] S. Maldonado Bascon, J. Acevedo
Rodriguez, S. Lafuente Arroyo, A. Fernández
Caballero, F. Lopez-Ferreras, “An optimization

IJMSRT25APR044 www.ijmsrt.com 199

DOI: https://doi.org/10.5281/zenodo.15300745