Conference PaperPDF Available

Traffic Sign Recognition System (TSRS): SVM and Convolutional Neural Network


Abstract and Figures

TSRS (Traffic Sign Recognition System) may play a significant role in the self-driving car, artificial driver assistance, traffic surveillance as well as traffic safety. Traffic sign recognition is necessary to overcome the traffic-related difficulties. The traffic sign recognition system consists of two parts—localization and recognition. In the localization part, traffic sign region is located and identified by creating a rectangular area. After that, in recognition part, the rectangular box provided the result for which traffic sign is located in that particular region. In this paper, we describe an approach toward the traffic signs recognition system. Here, we worked on 12 selected signs for traffic sign detection and recognition purpose. In this intention, we used a support vector machine (SVM) and convolutional neural network (CNN) individually to detect and recognize the traffic signs. We obtained 98.33% accuracy for SVM with an 80:20 train and test data ratio. On the other hand, the test result was 96.40% accurate for CNN.
Content may be subject to copyright.
Traffic Sign Recognition System (TSRS): SVM and
Convolutional Neural Network
Nazmul Hasan1, Tanvir Anzum1, and *Nusrat Jahan2
1Computer Science and Engineering Department
Daffodil International University
{nazmul15-7914, tanvir15-7890}
2Senior Lecturer
Computer Science and Engineering Department
Daffodil International University
Abstract. TSRS (Traffic Sign Recognition System) may plays a significant
role in self driving car, artificial driver assistances, traffic surveillance as
well as traffic safety. Traffic sign recognition is necessary to overcome the
traffic related difficulties. The traffic sign recognition system has two parts-
localization and recognition. In localization part, where traffic sign region is
located and identified by creating a rectangular area. After that, in
recognition part the rectangular box provided the result for which traffic
sign is located in that particular region. In this paper, we describe an
approach towards traffic signs recognition system. Here, we worked on 12
selected sign to traffic sign detection and recognition purpose. In this
intention, we used Support Vector Machine (SVM) and Convolutional
Neural Metwork (CNN) individually to detect and recignize the traffic
signs. We obtained 98.33% accuracy for SVM with 80:20 train and test data
ratio. On the other hand, the test result was 96.40% accurate for CNN.
Keywords: Image processing. SVM. CNN. traffic sign.
1 Introduction
With the rapid growth of technological development, vehicles have become an
essential portion of in our routine lives. Because driving vehicles without follow
traffic rules, it creates more and more intricate traffic on the road. As a result, it is
one of the major reasons behind accidents every year. In recent times road
accidents are happening regularly in increasing manner across the world. Leading
reason of most road accidents is the ignorance or unawareness of the traffic sign.
The meaning of traffic sign is any entity, device, or board on the road that entity
carries the rules, indicates the warning or provides other explanation regarding
driving. Therefore, it also provides necessary information through traffic signals
and traffic control devices to continue smooth car driving.
Traffic sign detection and recognition system is an crucial issue to reduce taffic
and increase the sustainability of selfdriving car without any incidence. TSRS
plays a crucial role in autonomous vehicle, smart driving and smart traffic system.
The traffic sign is used in Bangladesh since 1930’s and those are inadequate in
this current traffic situations. So, we need to underpin TSRS to provide an
autonomous vehicle to reduce traffic as well as road accident.
An approach assemble with video camera and an active machine with the
vehicle is a simple driver assistance system focused on frame by frame
observation of the motion frames can be implemented and there by generated the
alert signals accordingly. So, the driver would be able to take the decision
effortlessly. Road sign analysis is one of the significant aspects for automatic
driving support system. Real time qualitative road sign observation is the root for
any updated transport system. In this paper, we proposed this system to recognize
traffic sign with supervised classification algorithms. The outcome of this study
can be used for recognizing traffic signs and direction also to slow down or direct
driver to another safe route.
Fig. 1. Basic block diagram for TSRS
A simple block diagram to present the overall prospective has depicted in
figure 1. With the help of this diagram, we demonstrated the proposed system
towards automated car. In our proposed approach images is captures from vehicle
using a video camera. After pre-processing input data classification algorithm was
established to detect and recognize the traffic signs. Finally, we introduce a driver
alert system to minimize the error of car driving approach.
2 Related Work
This section for highlights some recent related research works. Autonomas car is a
recent research topic. We found many researches on this field and the working
progress is too high.
In 2019, Wei-Jong Yang et al. proposed an approach to recognize traffic sign.
They worked with shaped based detection algorithms and for classification
purpose they choose convolutional neural network. After simulation they got 97%
sign recognition accuracy [1]. In this paper [2] author proposed SVM based
classification algorithm to recognize traffic sign. Here, they considered 8 types of
road signs. For training purpose they used 600 different images for each signs and
for test purpose 120 images was considered. In this paper, they tested individual
signs with real data and their accuracy level was 66.6% to 100%. Prashengit Dhar
et al. in 2017 proposed a Traffic Sign Recognition (TSR) system. Here, they were
used HSV color model and deep CNN for automatic features extraction as a
classifier. After this study, they achieved 97% accuracy [3]. In 2019 [4] Aashrith
and et al. used Convolutional Neural Networks for recognizing traffic sign. Here,
they found 99.18% accuracy on using Belgium Data and German Traffic Sign
Benchmark (GTSDB). Where as they got 99.50% accuracy almost 0.32%
improvement at accuracy. In 2016 at paper [5], Di Zang et al. classified their
dataset using Support Vector Machine then detection part is done by CNN. Their
accuracy was almost 96.50%, and they used GTSDB dataset. In 2017 at paper [6],
Ardianto et al. used SVM to classified object and got 91.5% accuracy in detection
purpose and they also used GTSDB dataset. They improved it to add a feature
called Histogram of Oriented Gradients (HOG) that help to increase its accuracy
up to 98%. In 2017 Shi et al., applied SVM to detect which region of image
contain a traffic sign [7].
Pavly Salah Zaki and et al. worked on traffic sign detection, multiple entity
detection system. In 2019, they used Faster Recurrent CNNs, Single Shot Multi
Box Detector along with several feature extractors to detect traffic sign. However,
they underpin F-RCNN to get best results. Here, they used the GTSDB dataset.
The GTSDB holds complete 900 images, where 800 for training and 100 for
testing [8]. In 2020, Yanmei Jin and et al. also worked with GTSRB dataset to
propose a Single Shot Detector algorithm combine with multi feature fusion and
they called it MF-SSD. However, in this time they divided total 900 images into
600 training and 300 as a test data to detect traffic sign [9]. In 2017, Yassmina
Saadna and Ali Behloul discussed a system to identify and recognize traffic sign.
Their main goal was to find out traffic sign detection methods to locate the
regions-of-interest that contain traffic sign. They divided the methods into 3 steps-
focused color, shape, and finally considered learning based methods [10]. In 2016,
for detecting and classifying the traffic signs they proposed an approach. It has 2
main steps: road sign detection, after that classification with recognition. To
classify the traffic sign they used neural network and to complete this work they
picked four types of traffic signs: Stop, No Entry, Give Way, and Speed Limit
Sign. Considered total 3 hundreds sets images, and they got 90% and 88%
accuracy for detection and recognition purpose [11]. So, detection and recognition
of traffic sign is necessary to build an autonomas car driving system.
3 Proposed Methodology
To recognition traffic sign we focused two machine learning algorithms. Many
recent research works on traffic sign used SVM and CNN. For detecting and
recognizing traffic sign we used these two algorithms. In the next step we also
evaluated the proposed approach with CNN and SVM.
3.1 Data Collection and Preparation
To complete this study a dataset was built from cropping video frame. We also
collect some random video and crop traffic sign area to build a real dataset. Then,
we categorized its own classes and split the whole data into training and validation
dataset. We have total 1200 images to propose SVM and CNN model.
Table 1. 12 Traffic Signs with Description.
Traffic Sign
Traffic Sign
Turn Left
Only Left
Only Right
Turn Right
Road Merges
Speed Breaker
We have considered 12 different classes and each class completed with 100
images. Split the whole dataset as training and validation purpose. To generate a
model with SVM, 80% data for trining and 20% data for testing purpose is
considered. Meanwhile, for CNN classification we have divided dataset into two
classes train and validation. For training purpose data volume was 1080 images
and 120 images for testing purpose, where total 12 different traffic signs was
contained. Table 1 to represnt the selected 12 traffic signs. Here, we illstrated evey
traffic sign with sample an image.
In the next section, we are attending to elaborate the work procedure of our
proposed approach to detect and recognize traffic sign.
3.2 Convolutional Neural Network
Convolutional neural network is a part of deep learning approch. Convolutional
neural network represents a vast breakthrough in image recognition. Many
research fields focused on CNN to achieve the highest accurate result. A CNN
model constracts with following four layers:
Convolutional layers
Relu function
Polling layers
Fully connected layer
Convolution is the first step to create complete network. Here, Convolutioal
operates on two images in 2D format. One as a input image, and another output
image. It helps to understand the features from images by making relation between
pixels. A convolution of two funcation f and g is defined with following the
(𝑓 𝑔)(𝑖) = ∑ 𝑔(𝑗) 𝑓(𝑖 𝑗 +m/2) (1)
Next, relu is placed after convolution operation. This “relu” fuction considered
for executing non-linear operation. It performs on pixel wise operation and sets 0
insted of all negative values. There are other non-linear functions such as tanh or
sigmoid also used in CNN but “relu” performs best amon them in different
situations. On the other hand, Polling progressively decrese the size of the input
images. Polling helps to decrese the number of dimentions and the number of
claculation required. Polling also helps to control overfitting. There are many
types of polling. But in this paper we used max polling.
Fully connected layer, combines our features into output stage. It predicts the
class level with better accuracy. The error is computed and then occured back
propagation. This process is continue layer by layer. A sample flowchart for our
proposed model on CNN and SVM is given in figure 2.
Fig. 2. Flowchart of our model
3.3 Support Vector Machine (SVM)
SVM is a familiar supervised machine learning algorithm. It is usually worked for
regression and classification problems. This algorithm distinct classess by
generating hyperlane between two classes. Actually it divides dataset into required
classes. In the SVM model, it will be clearly understanding with an example for
two classes. We can easily understand the process of SVM model with the help of
figure 3.
Fig. 3. Spilt data into two classes (SVM)
To understand the work process of SVM we neeed to know the region-of-
interest “ROI module. This region of interest follow three steps. First one is,
color transformation. That converts the colourful images into grayscale images.
The second step is to control shape of images. Finally, it refines the ROI. This
exploits the regularity of traffic signs with their color and shape with better
In this paper, to describe the methodology, the size of the cells and blocks are
varied to get several image sizes. We used 9-bin histograms for this methodology.
To train the model we rescale the training images to 32x32 pixel images. HOG
descriptors to calculate and train SVM classifier. The m number of support vectors
are merged to a single one where multiply every vector with its weight “a” and
adding it with a global vector “v”. Equation 2 for cumputing the suport vector
𝑣 = ∑ 𝛼𝑖. 𝑣𝑖 (2)
Now, we are going to illustrate the whole work process of SVM with the help of
figure 4. Here, we depict all the steps of SVM to reciognize the traffic signs for
Fig. 4. SVM for Traffic Sign Detection and Recognition
4 Result Analysis
In this section we are going to discuss about result which was obtained from SVM
and CNN. SVM classification for Traffic sign detection and recognition provided
98.33% accuracy (80:20 data split ration), while in CNN method we achieved
99.56% training accuracy and 96.40% validation accuracy. The training accracy,
validation accuracy and loss of CNN model is visualized in figure 5 and 6.
Fig. 5. Accuracy of Training and validation
Fig. 6. Line chart for Training and validation loss
For SVM method we were getting 98.33% accuracy when we splited total
dataset into 80:20 ration for training and testing purpose. However, we acieved
99.17% accuracy when considered 90:10 ration. Figure 7 for presenting the
performance of SVM.
Fig. 7. Result of SVM
After that, we developed a system using SVM to evaluate the results with real
time video. The detection part uses image processing techniques that creates
contour on each frame and finds all ellipse or circles among those contours. Then,
the detection part marked as categorized traffic sign. Some sample results shown
in following figures.
Fig. 8. Recognize Danger Sign
Fig. 9. Recognize Drive 30 km/h” Sign
Fig. 10. Recognize Turn Right” Sign
Figure 8, 9, and 10 to understand the performance of SVM with the help of some real
time data. If we implement the full system with CNN then it will be work well then SVM
as we have mentioned their accuracy level. After complete our study we obtained TSRS can
be an established and sustainable system. Many researches are going on with traffic issues
and self-car driving purpose. We needed an accurate model to implement this system to
provide the better solution in every case.
5 Conclusion
This paper study was to represent an original effective traffic sign detection and
recognition approach towards the design of TSRS. As a recent research topic
TSRS is getting popular day by day. In this study, it is done using SVM and
CNN classification algorithms to decline extensive traffic difficulties. In our ex-
periment, we obtained highest training accuracy from CNN 99.56%, while the
test accuracy was 96.40%. We showed the real time evaluation results of SVM,
where the system performed 98.33% accurately. Many research focused on SVM
and CNN to solve this specific problem.
In future, our aim is to increase the number of traffic sign classes with large
amount of quality data. As in a machine learning research, to maintain data vol-
ume and data quality is most important and time consuming part. To provide a
complete system to overcome the traffic issues our ambition is to implement a
system with distance calculation form car to traffic sign.
1. Yang, WJ., Luo, CC., Chung, PC., Yang, JF.: Simplified Neural Networks with
Smart Detection for Road Traffic Sign Recognition. Lecture Notes in Networks
and Systems, vol 69. Springer, Cham (2020)
2. Dubey, A.R., Shukla, N., Kumar, D.: Detection and Classification of Road Signs
Using HOG-SVM Method. Smart Computing Paradigms: New Progresses and
Challenges. Advances in Intelligent Systems and Computing, vol 766. Springer,
Singapore (2020)
3. Dhar, M. Z., Abedin, T., Biswas, Datta, A.: Traffic sign detection A new
approach and recognition using convolution neural network. IEEE Region 10
Humanitarian Technology Conference (R10-HTC), Dhaka, pp. 416-419, (2017)
4. Aashrith, V., Smriti, S.: Traffic Sign Detection and Recognition using a CNN
Ensemble. IEEE International Conference on Consumer Electronics (ICCE). 07
March, (2019)
5. Di, Z., Junqi, Z., Dongdong, Z.: Traffic sign detection based on cascaded
convolutional neural networks. In: Conference on Software Engineering,
Artificial Intelligence, Networking and Parallel/Distributed Computing
(SNPD), (2016)
6. Sandy, A., Chih-Jung, C., Hsueh-Ming H.: Real-time traffic sign recognition
using color segmentation and SVM. In: Conference on Systems, Signals and
Image Processing (IWSSIP), (2017)
7. Jian-He, S., Huei-Yung, L.: A vision system for traffic sign detection and
recognition. IEEE 26th In: Symposium on Industrial Electronics (ISIE), (2017)
8. Zaki, P. S., et al.: Traffic Signs Detection and Recognition System using Deep
Learning. In: Conference on Intelligent Computing and Information Systems
(ICICIS) (2019)
9. Jin, Y., Fu, Y., Wang, W., Guo, J., Ren, C., Xiang, X.: Multi-Feature Fusion and
Enhancement Single Shot Detector for Traffic Sign Recognition. in IEEE Access,
vol. 8, pp. 38931-38940, (2020)
10. Saadna, Y., Behloul, A.: An overview of traffic sign detection and classification
methods. In: J Multimed Info Retr 6, 193210.
017-0129-8 (2017)
11. Sheikh , M. A. A., Kole , A., Maity, T.: Traffic sign detection and classification
using colour feature and neural network. In: Conference on Intelligent Control
Power and Instrumentation (ICICPI), Kolkata, pp. 307-311. (2016)
12. Fatin, Z., Bogdan, S.: Warning traffic sign recognition using a HOG-based K-d
tree. IEEE Intelligent Vehicles Symposium (IV), (2011)
The traffic signal recognition model plays a significant role in the intelligent transportation model, as traffic signals aid the drivers to driving the more professional with awareness. The primary goal of this paper is to proposea model that works for the recognition and detection of traffic signals. This work proposes the pre-processing and segmentation approach applying machine learning techniques are occurred recent trends of study. Initially, the median filter & histogram equalization technique is utilized for pre-processing the traffic signal images, and also information of the figures being increased. The contrast of the figures upgraded, and information about the color shape of traffic signals are applied by the model. To localize the traffic signal in the obtained image, then this region of interest in traffic signal figures are extracted. The traffic signal recognition and classification experiments are managed depending on the German Traffic Signal Recognition Benchmark-(GTSRB). Various machine learning techniques such as Support Vector Machine (SVM), Extreme Learning Machine (ELM), Linear Discriminant Analysis (LDA), Principal Component Analysis (PCA), Convolutional neural network (CNN)- General Regression Neural Network (GRNN) is used for the classification process. Finally, the obtained results will be compare in terms of the performance metrics like accuracy, F1 score, kappa score, jaccard score, sensitivity, specificity, recall, and precision. The result shows that CNN-GRNN with ML techniques by attaining 99.41% accuracy compare to other intelligent methods. In this proposed technique is used for detecting and classifying various categories of traffic signals to improve the accuracy and effectiveness of the system.
Full-text available
The computer plays a significant role in computer vision to help with everyday tasks. Defense, biometrics, visual surveillance, robotics, and driver assistance are just a few of the many uses for object recognition. Lane/railroad track detection, the detection of obstacles before the vehicle/train, is part of the driver assistance system. Driver assistance systems for various modes of transportation can be improved by employing an efficient object-recognition approach (road, rail, etc.). One of the major problems with the driver assistance system is its inability to detect railroad obstacles. There has been an increase in the amount of research and development in obstacle detection for road transportation in the last few years. Despite the fact that railroads are the other primary land transportation mode, much less effort has been put into developing technologies for detecting obstacles on the rails than on the roads. Efforts to improve the recognition performance of safety inferences are ongoing. Real-time object recognition in driving situations despite the rapid development in the field of object recognition on datasets with a tremendous number of different types of objects remains extremely challenging. Autonomous driving systems (ADAS) and advanced driver assistance systems (ADAS) face a number of key difficulties in visual object recognition. Object recognition is complicated by a variety of factors, including changes in lighting conditions, the presence of shadows, and partial occlusion, for example. Changing light conditions are the most common triggers for variation. These critical realities are taken into account in this study, which aims to offer a solution to these problems.
Full-text available
Automatic Traffic Sign Detection and Recognition (TSDR) provides drivers with critical information on traffic signs, and it constitutes an enabling condition for autonomous driving. Misclassifying even a single sign may constitute a severe hazard, which negatively impacts the environment, infrastructures, and human lives. Therefore, a reliable TSDR mechanism is essential to attain a safe circulation of road vehicles. Traffic Sign Recognition (TSR) techniques that use Machine Learning (ML) algorithms have been proposed, but no agreement on a preferred ML algorithm nor perfect classification capabilities were always achieved by any existing solutions. Consequently, our study employs ML-based classifiers to build a TSR system that analyzes a sliding window of frames sampled by sensors on a vehicle. Such TSR processes the most recent frame and past frames sampled by sensors through (i) Long Short-Term Memory (LSTM) networks and (ii) Stacking Meta-Learners, which allow for efficiently combining base-learning classification episodes into a unified and improved meta-level classification. Experimental results by using publicly available datasets show that Stacking Meta-Learners dramatically reduce misclassifications of signs and achieved perfect classification on all three considered datasets. This shows the potential of our novel approach based on sliding windows to be used as an efficient solution for TSR.
Full-text available
Smart electric vehicles (EVs) are attractive because of their clean, zero-emission, low impact on the environment whilst providing a safer and smoother riding experience. To provide the latter, driving control requires appropriate systems and algorithms to optimize smart vehicle performance, maximize vehicle stability and protection, minimize accident probability, heighten driving comfort, and optimize transportation costs. Despite advancements in these areas, the realization of optimal smart EVs still requires considerable effort. This paper reviews driving control systems and algorithms for smart EVs, including the advanced driving assistant system, implementation of sensors, vehicle dynamics, and control algorithms. The major contribution of this review is to identify promising work to assist researchers with the most advanced trends in this area for prospective regulations.
Automatically identifying traffic signs is a challenging and time-consuming process. As the academic community pays more attention to traditional algorithms for vision-based detection, tracking, and classification, three main criteria drive the investigation, they are detection, tracking, and classification. It is capable of performing detection and identification operations to minimize traffic accidents and move towards autonomous cars. A novel method proposed in this paper is based on moment invariants and neural networks for performing detection and recognition with classification, and it also includes automatic detection and identification of traffic signs and traffic board text that uses colour segmentation. Aside from the proposed structure, it is also required to identify the potential graphic road marking with text. This research article contains two algorithms, which are used to accurately classify the board text. The detection through image segmentation and recognition can be done by using the CNN algorithm. Finally, the classification is performed by the SVM framework. Therefore, the proposed framework will be very accurate and reliable with high efficiency, which has been proven in many big dataset applications. The proposed algorithm is tested with various datasets and provided good identification rate compared to the traditional algorithm.
Full-text available
Road traffic sign detection and recognition play an important role in advanced driver assistance systems (ADAS) by providing real-time road sign perception information. In this paper, we propose an improved (Single Shot Detector) SSD algorithm via multi-feature fusion and enhancement, named MF-SSD, for traffic sign recognition. First, low-level features are fused into high-level features to improve the detection performance of small targets in the SSD. We then enhance the features in different channels to detect the target by enhancing effective channel features and suppressing invalid channel features. Our algorithm gets good results in domestic real-time traffic signs. The proposed MF-SSD algorithm is evaluated with the German Traffic Sign Recognition Benchmark (GTSRB) dataset. The experimental results show that the MF-SSD algorithm has advantages in detecting small traffic signs. Compared with existing methods, it achieves higher detection accuracy, better efficiency, and better robustness in complex traffic environment.
Full-text available
Over the last few years, different traffic sign recognition systems were proposed. The present paper introduces an overview of some recent and efficient methods in the traffic sign detection and classification. Indeed, the main goal of detection methods is localizing regions of interest containing traffic sign, and we divide detection methods into three main categories: color-based (classified according to the color space), shape-based, and learning-based methods (including deep learning). In addition, we also divide classification methods into two categories: learning methods based on hand-crafted features (HOG, LBP, SIFT, SURF, BRISK) and deep learning methods. For easy reference, the different detection and classification methods are summarized in tables along with the different datasets. Furthermore, future research directions and recommendations are given in order to boost TSR’s performance.
Artificially intelligent systems are becoming a crucial part of the future we are heading to. In this paper, we aim to produce an intelligent software system that can recognize the traffic warning signs. This paper also describes a road sign recognition (RSR) system, a step toward driverless cars. It uses the color-based localization and classification capabilities of support vector machines over the features described as histogram of oriented gradients. The proposed system is exhaustively tested on a dataset of realistic images. The obtained encouraging results validate the potential of the proposed method.
Improving driver’s safety is the main goal of the advanced driver assistance system, which has been widely deployed for proactive driving security in recent years. For road driving, the advanced driver assistance system should visually recognize circular prohibition and triangular warning traffic signs to help drivers to grab complete traffic conditions. In this paper, we proposed a low-computation neural assistance system for traffic sign recognition. First, we proposed shaped-based detection algorithms to detect the regions, which are with circle and triangular traffic signs in designated regions of interest. For classification to those detected regions, we then suggest a convolutional neural network to achieve about 5% improvement of top 1 accuracy compared with LeNet model in German traffic sign recognition benchmarks dataset. For real applications, we also establish a Taiwanese traffic sign database to train the proposed neural network. The simulation results on self-collect driving videos demonstrate that the proposed traffic sign recognition system achieved above 97% recognition rate can be effectively adopted in ADAS applications.
Conference Paper
Automatic traffic sign detection and recognition is a field of computer vision which is very important aspect for advanced driver support system. This paper proposes a framework that will detect and classify different types of traffic signs from images. The technique consists of two main modules: road sign detection, and classification and recognition. In the first step, colour space conversion, colour based segmentation are applied to find out if a traffic sign is present. If present, the sign will be highlighted, normalized in size and then classified. Neural network is used for classification purposes. For evaluation purpose, four type traffic signs such as Stop Sign, No Entry Sign, Give Way Sign, and Speed Limit Sign are used. Altogether 300 sets images, 75 sets for each type are used for training purposes. 200 images are used testing. The experimental results show the detection rate is above 90% and the accuracy of recognition is more than 88%.