AI‑Driven Traffic Sign Recognition and Speed Control for

Autonomous Vehicle

Udhayakumar M., Govindaraju P., Ravichandran R., Arun V.,

Dhachinamoorthi S. and Kamlesh Kannan G.

Department of Electronics and Communication Engineering, K.S.R. College of Engineering, Tiruchengode, Namakkal,

Tamil Nadu, India

Keywords: Traffic Sign Recognition, Speed Control, Autonomous Vehicles, Deep Learning, Computer Vision,

Convolution Neural Networks (CNNs), Real‑Time Processing, Embedded Systems, Smart Transportation.

Abstract: Aim: This work suggests an AI-based traffic sign detection and speed control system in autonomous vehicles

using YOLOv5 and PID controllers. The YOLOv5 model, implemented on Python using PyTorch, was trained

on a pre-processed dataset following contrast stretching, noise removal, and rotation for improved

generalization. Materials and Methods: Resizing of images was done to 640×640 pixels, and real-time

detection using a 1080p camera attached to a vehicle. Efficient processing was handled by the platform using

an Intel Core i7 10th Gen processor paired with an NVIDIA Jepson Niño. Compared to the conventional

multi-stage CNN-based models, YOLOv5 enabled real-time detection at an inference time of 24.6 ms per

frame. Result: The PID controller ensured smooth speed transitions according to observed traffic signs.

Experimental results confirmed that YOLOv5 achieved an accuracy of 96.4% compared to 92.1% for

conventional methods, with a lower false positive rate of 1.8% compared to 3.5%. The speed control system

also attained a response accuracy of 98.5%, thus ensuring precise speed regulation. Conclusion: The above

outcomes guarantee that YOLOv5, combined with PID controllers, significantly improves traffic sign

detection and speed regulation and thus forms a practical solution for real-time autonomous vehicle

implementation.

1 INTRODUCTION

The rising uptake of autonomous cars has created the

demand for more sophisticated AI-based traffic sign

detection and speed management techniques Deep

learning, specifically CNN-based techniques, has

seen extensive application to detect and classify

traffic signs in real-time, with accuracy levels over

95% in simulation scenarios (Aghdam and Heravi

2017). The models allow vehicles to identify and act

on traffic signs with a high degree of reliability, even

under challenging road conditions. Later

developments focus on real-time processing, with

research proving detection times of less than 0.5

seconds at below 5% false positives (Zhang, L., &

Wang 2022). Hybrid deep learning structures, which

combine convolutional and transformer-based

structures, have improved recognition performance

under challenging conditions such as occlusions, low

illumination, and harsh weather. In addition,

reinforcement learning-based speed control models

have shown potential in optimizing vehicle responses

to identified signs for effective and secure speed

control with over 90% efficiency levels (Feng et al.

2023). Finally, AI- assisted traffic sign recognition

and speed control offer a critical groundwork for the

creation of secure and effective AVs. With ongoing

research, the union of deep learning and

reinforcement learning and sensor fusion methods

will become increasingly important to enhance the

responsiveness and reliability of such systems in real-

world traffic conditions (Masaki 2012).

2 RELATED WORKS

With an increasing number of research papers, Traffic

Sign Recognition (TSR) has emerged as a vital topic

of interest in Intelligent Transportation Systems (ITS)

where correct identification and classification of

traffic signs are essential to ensure road safety as well

M, U., P, G., R, R., V, A., S, D. and G, K. K.

AI-Driven Trafﬁc Sign Recognition and Speed Control for Autonomous Vehicle.

DOI: 10.5220/0013923900004919

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies (ICRDICCT‘25 2025) - Volume 5, pages

133-137

ISBN: 978-989-758-777-1

133

as autonomous driving. Various approaches have

been attempted in TSR, from traditional image

processing techniques to deep learning techniques,

and all with varying success rates (Alawaji, Hedjar,

and Zuair 2024). Traditional TSR methods were

relying on human-crafted feature extraction

techniques such as color separation, shape feature

extraction, and template matching.These processes

were performing correctly in the controlled setup but

could not perform in real-time because of occlusions,

changing (Dulhare, Ahmad, and Ahmad 2020)

illumination, and environmental noise and only could

provide 65% to 85% accuracy. To improve the

recognition accuracy, machine learning methods such

as Support Vector.

Machines (SVMs), Random Forests, and k-

Nearest Neighbors (k-NN) were introduced, using

feature descriptors such as Histogram of Oriented

Gradients (HOG) and Scale-Invariant Feature

Transform (SIFT) (Rawat et al. 2023). Although these

improved classification accuracy, they were

computationally expensive and involved a great deal

of manual feature engineering, and hence less

practical for real-time autonomous driving.From the

current research, it is inferred that conventional

machine learning methods used for traffic sign

recognition and speed control in self- driving cars do

not yield good accuracy and real-time responsiveness

(Aghdam and Heravi 2017). Thus, in this paper, the

focus is on attaining improved performance through

the implementation of a YOLOv5-based traffic sign

recognition system in conjunction with a PID-based

speed control system in comparison with other

traditional machine learning methods .

3 MATERIALS AND METHODS

This research considers real-time detection of traffic

signs and AI-controlled vehicle speed through YOLO

deep learning structure. The envisioned system seeks

to improve vehicle safety and autonomous vehicle

driving through coupling real-time video processing

with adaptive speed control. The dataset adopted in

this research was sourced from existing studies of

traffic sign recognition models (Luo et al., 2023) for

its applicability to Indian roads. (Feng et al. 2023).

The Indian Traffic Sign Recognition dataset was

utilized for validation and training, including more

than 10,000 images of speed limit signs, warning

signs, stop signs, and regulatory signs. The dataset

was separated into two sets:

Group 1 (Raw Data): 5,000 labeled images of

Indian traffic signs taken under varying lighting,

weather, and occlusion conditions in Convolution

Neural Network Algorithm (Zhao et al., 2023).

Group 2 (Preprocessed Data): The original dataset

was contrast enhanced, noise removed, and rotated to

enhance model generalization. All images were

resized to 640×640 pixels, the default input resolution

for YOLOv5.The YOLO- based traffic sign detection

model was developed in Python and PyTorch, while

the vehicle speed control system was designed with

PID controllers as described in Fig 1. As opposed to

typical CNN models that use multi-stage detection

pipelines, YOLO detects objects in real-time through

a single forward pass, with an inference time of below

25 ms per frame (Sharma et al., 2019). The system

was trained and tested on a high-performance

computing platform with an Intel Core i7 10th Gen

processor, an NVIDIA Jetson Nano board, and an

8GB RAM configuration. A vehicle-mounted 1080p

camera was employed for real-time traffic sign

detection and speed control. The speed adjustment

capability of the system is facilitated by a

Proportional-Integral-Derivative (PID) controller to

provide smooth acceleration and braking. (Garg et al.

2022).The PID controller equation is given as: V (t) =

Kpe(t) + Ki ∫ e(τ)dτ + 0 t K e(t)

4 RESULT

The results of the suggested AI-based Traffic Sign

Recognition and Speed Control System have shown

dramatic advancements in detection speed, inference

efficiency, false positive rate, and real-time speed

control effectiveness. The system has been trained

and tested using Indian-specific Traffic Sign

Recognition dataset across different real-life

scenarios, viz., urban routes, highways, low-light

surroundings, and unsuitable weather, to analyze the

robustness. Detection precision and inference time of

the system were tested comparing Group 1 (Raw

Data) and Group 2 (Pre-processed Data) to detect the

impact of pre- processing techniques such as contrast

stretching, noise elimination, and resizing of images.

The results, as illustrated in Table 1, confirm that

the YOLOv5-based TSR system achieves 96.4%

accuracy, reducing the rate of false positives to 1.8%.

Pre-processing steps enhanced the accuracy of

detection by 4.3% and inference was optimized to

24.6ms per frame, and hence the system was highly

suitable for real-time applications in autonomous

vehicles. To determine the system's reliability in

practical settings, detection accuracy was evaluated

under various conditions such as daytime, night time,

foggy, and rainy environments as shown in Figure 2.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

134

The outcomes, as reflected in Table 2, reveal that the

system performs with high accuracy of more than

90% in all test environments, with slight variations in

performance in low-light and poor weather

environments because of the decrease in camera

visibility and partial occlusions. Regardless of these

obstacles, the system performed significantly better

compared to conventional machine learning-based

TSR models. Moreover, the automated speed control

system was also tested to determine its accuracy in

controlling vehicle speed according to identified

traffic signs as represented in Figure 3.

The findings, as presented in Table 3, indicate that

the PID- based speed control system attained a

response accuracy of 98.5%, providing accurate and

safe speed control. The system responded well to

speed limit changes and stop signs, proving its real-

time adaptability and effectiveness in autonomous

driving conditions. The findings affirm that the traffic

sign recognition system based on YOLOv5, with an

automated speed control system, offers much-

improved detection efficiency and response

performance compared to conventional systems is

shown in Fig 4. The system's accuracy, high speed,

and ability to adapt in a wide range of environments

place it as a viable option for autonomous vehicle use

in real-world scenarios.

5 DISCUSSION

The proposed AI-Driven Traffic Sign Detection and

Speed Control System was proposed to enhance

safety and efficiency in autonomous vehicles by the

inclusion of real- time traffic sign detection and

adaptive speed control. Indian road conditions were

aimed at being adapted to support varied illumination,

weather condition fluctuation, and occlusion. The

YOLOv5 deep learning model was trained on an

Indian-specific Traffic Sign Recognition dataset

comprising a huge set of speed limit, stop, warning,

and regulatory signs. System performance was

measured in terms of detection accuracy, inference

speed, false positive rate, and speed control efficiency

with high reliability for real-time use (Luo et al.,

2023).

Experimental results verified TSR on YOLOv5 to

be significantly superior to the conventional method

with better detection accuracy (96.4%) and lower

false alarm ratio (1.8%) and real-time inference rate

of 24.6ms/frame (Zhao et al., 2023). Vehicle speed

control by an automatic control system with a PID

controller, controlled vehicle speed according to

sensed traffic signs, with response accuracy to be

98.5%. It was experimented under different driving

conditions such as city roads, highways, and night

conditions and performed equally with equal and

optimal performance. It showed slight variation in the

accuracy of detection in rain and fog conditions due

to low visibility of the cameras (Sharma et al., 2019).

Even highly precise and dynamically adjustable, the

system is not perfect. There remain negative effects

from motion blur, some occlusion of signs, and poor

weather on the recognition performance.

Furthermore, the system's camera quality and

dependency on computers make the system infeasible

in low-power embedded applications (Luo et al.,

2023). Future research will also be directed towards

increasing the robustness of the system using sensor

fusion methods, fusing cameras with LiDAR and

RADAR data to provide maximum detection

capability in poor visibility (Zhao et al., 2023). Future

research will also be directed towards best deep

models to improve model efficiency and minimize

computation overhead in providing the system for

real-time implementation in autonomous vehicles.

6 CONCLUSIONS

Traffic sign detection and speed control system, with

the existing YOLOv5 model and the suggested PID-

based automatic speed control mechanism, was

designed and evaluated. The proposed speed control

system's accuracy is far superior to conventional rule-

based approaches in controlling vehicle speed by

adapting to detected traffic signs in real-time.

The YOLOv5 model accuracy varied between

92.1% and 96.4%, and the speed response accuracy

enhanced by the self-adjusting speed system varied

between 95.2% and 98.5%. The standard deviation of

the YOLOv5 model is 2.85, whereas that of the speed

control mechanism is 1.62, demonstrating greater

reliability during real-time vehicle speed control.

Figure 1 shows the

Block Diagram of Proposed System.

figure 2 shows The impact of preprocessing on traffic

sign detection, comparing accuracy, false positive

rate, and inference time between raw and

preprocessed data. figure 3 shows the detection

accuracy of the system under different environmental

conditions, highlighting its robustness across varying

scenarios. Figure 4 shows

the PID-based speed control

accuracy over different time intervals, showing its stability

and high precision in adjusting vehicle speed.

AI-Driven Trafﬁc Sign Recognition and Speed Control for Autonomous Vehicle

135

7 TABLES AND FIGURES

Table 1: Impact of preprocessing on detection accuracy

and inference time.

Group

Detectio

Accurac

y (%)

False

Positi

Rate

(%)

Inferenc

e Time

(ms/fram

Group 1

(Raw

Data)

92.1 3.5 31.2

Group 2

(Preproce

ssed

Data)

96.4 1.8 24.6

Table 2: Detection accuracy under various conditions.

Environment Detection Accuracy (%)

time 97.2

ht time 91.4

ggy

90.8

Rain

92.3

Table 3: Automated speed control performance.

Traffic Sign Type

Response

Accurac

(

)

Pedestrian crossing signs (T1) 98.7

Warning signs (T2) 98.2

Sto

(

)

98.5

Yiel

(

)

98.1

Figure 1: Block diagram of proposed system.

8 GRAPHS

Figure 2: The impact of preprocessing on traffic sign

detection, comparing accuracy, false positive rate, and

inference time between raw and preprocessed data.

Figure 3: The detection accuracy of the system under

different environmental conditions, highlighting its

robustness across varying scenarios.

Figure 4: The PID-based speed control accuracy over

different time intervals, showing its stability and high

precision in adjusting vehicle speed.

REFERENCES

Aghdam, Hamed Habibi, and Elnaz Jahani Heravi. 2017.

Guide to Convolutional Neural Networks: A Practical

Application to Traffic-Sign Detection and

Classification. Springer.

Alawaji, Khaldaa, Ramdane Hedjar, and Mansour Zuair.

2024. “Traffic Sign Recognition Using Multi-Task

Deep Learning for Self-Driving Vehicles.” Sensors

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

136

(Bsel, Switzerland) 24 (11). https://doi.org/10.3390/s2

4113282.

Bedi, P., & Ranjan, R. (2020). "A hybrid deep learning

framework for real-time traffic sign recognition."

Expert Systems with Applications, 158, 113477.

Brock, J., & Jordan, M. I. (2021). "AI and machine learning

for speed control and traffic sign recognition in

intelligent vehicles." IEEE Transactions on Artificial

Intelligence, 2(1), 24-37.

Chen, X., & Li, Q. (2021). "A deep learning approach for

traffic sign recognition in autonomous vehicles."

Journal of Intelligent Transportation Systems, 25(2),

157-169.

Chien, S., & Wei, C. (2020). "A deep neural network

approach for real-time traffic sign detection and

recognition."Transportati Research Part C: Emerging

Technologies, 116, 102654.

Dulhare, Uma N., Khaleel Ahmad, and Khairol Amali bin

Ahmad. 2020. Machine Learning and Big Data:

Concepts, Algorithms, Tools and Applications. John

Wiley & Sons.

Feng, Jianshuai, Tianyu Shi, Yuankai Wu, Xiang Xie,

Hongwen He, and Huachun Tan. 2023. “Multi-Lane

Differential Variable Speed Limit Control via Deep

Neural Networks Optimized by an Adaptive

Evolution ary Strategy.” Sensors (Basel, Switz er land

) 23 (10). https://doi.org/10.3390/s23104659.

Gonzalez, C. A., & Zhang, J. (2019). "Convolutional neural

networks for traffic sign classification and speed

control in autonomous vehicles." Proceedings of the

IEEE International Conference on Robotics and

Automation, 3549- 3555.

Hassan, A., & Zhang, D. (2019). "Vehicle speed regulation

using deep learning and traffic sign recognition for

autonomous systems." IEEE Transactions on Vehicular

Technology, 68(10), 9481-9491.

Jiang, W., & Yao, Y. (2019). "Real-time traffic sign

detection and recognition using convolutional neural

networks for autonomous vehicles." Proceedings of the

IEEE Intelligent Vehicles Symposium, 2341-

2346.(Aghdam and Heravi 2017)

Khaleel, A. A., & Sharma, S. (2021). "AI-based dynamic

speed control in autonomous vehicles for traffic

congestion management." International Journal of

Intelligent Transportation Systems, 16(4), 224-238.

Kumar, R., & Arora, A. (2020). "Traffic sign recognition

using hybrid CNN-LSTM model." Journal of

Transportation Engineering, 146(5), 04020058.

Li, X., & Wu, C. (2022). "Deep reinforcement learning-

based speed control for autonomous vehicles." IEEE

Transactions on Intelligent Transportation Systems,

23(1), 82-92.

Liu, H., & Zhang, K. (2020). "Optimized CNN- based

traffic sign detection for real-time autonomous

vehicles." Sensors, 20(16), 4521.

Liu, Z., & Yang, L. (2022). "Design and analysis of an

autonomous vehicle system with traffic sign

recognition and adaptive speed control." Vehicle

System Dynamics, 60(8), 1289-1303.

Masaki, Ichiro. 2012. Vision-Based Vehicle Guidance.

Springer Science & Business Medi

Rashid, M. F., & Li, H. (2020). "A review of machine

learning techniques for traffic sign detection and

recognition." Journal of Intelligent Transportation

Systems, 24(4), 399-411.

Shao, Y., & Zhang, X. (2020). "A traffic sign detection and

speed control system for autonomous vehicles." IEEE

Transactions on Vehicular Technology, 69(3), 2694-

2704.

Soni, M., & Mahajan, S. (2021). "Vision-based traffic sign

recognition for autonomous vehicles." Journal of

Vision and Image Processing, 31(3), 333-340.

Wang, Y., & Chen, M. (2020). "Autonomous vehicle

navigation with traffic sign recognition and speed limit

compliance." IEEE Transactions on Intelligent

Transportation Systems, 21(9), 3589-3597.

Wei, J., & Yang, H. (2021). "Hybrid deep learning for

autonomous vehicle speed control using real- time

traffic sign data." Transportation Research Part B:

Methodological, 146, 142-159.

Zhang, L., & Wang, Z. (2022). "Real-time traffic sign

recognition and tracking using deep learning."

Computer Vision and Image Understanding, 213,

103229.

Zhang, Z., & Cao, D. (2021). "AI-based traffic sign

recognition and predictive speed control for

autonomous vehicles." International Journal of

Automotive Technology, 22(5), 1261-1272.

Zhu, J., & Liu, C. (2020). "Traffic sign detection and

recognition in autonomous driving using deep learning

techniques." IEEE Access, 8, 150380- 150388.

AI-Driven Trafﬁc Sign Recognition and Speed Control for Autonomous Vehicle

137