MINISTRY OF EDUCATION AND TRAINING HO CHI MINH CITY UNIVERSITY OF TECHNOLOGY AND EDUCATION GRADUATION THESIS MECHATRONICS ENGINEERING RESEARCH AND SIMULATE SPEED SIGN RECOGNITION SYSTEM IN SELF-DRIVING CAR INSTRUCTOR : Ph.VU QUANG HUY STUDENT: NGUYEN ANH HUY SKL012643 Ho Chi Minh City, January 2024 HO CHI MINH CITY UNIVERSITY OF TECHNOLOGY AND EDUCATION FACULTY FOR HIGH QUALITY TRAINING BACHELOR THESIS RESEARCH AND SIMULATE SPEED SIGN RECOGNITION SYSTEM IN SELF-DRIVING CAR NGUYEN ANH HUY Student ID: 18146022 Major: MECHATRONICS ENGINEERING Advisor: Ph. VU QUANG HUY Ho Chi Minh City, Jan 2024 THE SOCIALIST REPUBLIC OF VIETNAM Independence – Freedom– Happiness -------- Ho Chi Minh City, Day……. GRADUATION PROJECT ASSIGNMENT Student name: Nguyen Anh Huy Student ID: 18146022 Student name: Student ID: Student name: Student ID: Major: Mechatronics Engineering Class: 18146CLA Advisor: Ph. Vu Quang Huy Phone number: Date of assignment: _____________________ Date of submission: _____________ 1.
Project title: Research and simulate speed sign recognition system in self-driving car. Initial materials provided by the advisor: ___________________________________ 3. Content of the project: _________________________________________________ 4. Final product: ________________________________________________________ CHAIR OF THE PROGRAM ADVISOR (Sign with full name) (Sign with full name) 1 THE SOCIALIST REPUBLIC OF VIETNAM Independence – Freedom– Happiness -------- Ho Chi Minh City, Day…….
ADVISOR’S EVALUATION SHEET Student name: Nguyen Anh Huy Student ID: 18146022 Student name: Student ID: Student name: Student ID: Major: Mechatronics Engineering Project title: Research and simulate speed sign recognition system in self-driving car. Vu Quang Huy EVALUATION 1. Content of the project:. Approval for oral defense? (Approved or denied) .) Ho Chi Minh City, Day…….
ADVISOR (Sign with full name) 2 THE SOCIALIST REPUBLIC OF VIETNAM Independence – Freedom– Happiness -------- Ho Chi Minh City, Day……. PRE-DEFENSE EVALUATION SHEET Student name: Nguyen Anh Huy Student ID: 18146022 Student name: Student ID: Student name Student ID: Major: Mechatronics Engineering Class: 18146CLA Project title: Research and simulate speed sign recognition system in self-driving car. Vu Quang Huy Name of Reviewer:. Content and workload of the project.
Approval for oral defense? (Approved or denied) .) Ho Chi Minh City, Day……. REVIEWER (Sign with full name) 3 THE SOCIALIST REPUBLIC OF VIETNAM Independence – Freedom– Happiness -------- EVALUATION SHEET OF DEFENSE COMMITTEE MEMBER Student name: Nguyen Anh Huy Student ID: 18146022 Student name: Student ID: Student name: Student ID: Major: Mechatronics Engineering Class: 18146CLA Project title: Research and simulate speed sign recognition system in self-driving car. Vu Quang Huy Name of Defense Committee Member:. Content and workload of the project .) Ho Chi Minh City, Day…….
COMMITTEE MEMBER (Sign with full name) 4 ACKNOWLEDGEMENT This thesis was designed to examine speed limit sign recognition and how it impacts vehicle parameter adjustments. First of all, I would like to thank my thesis adviser Vu Quang Huy for his assistance, knowledge and persistent efforts towards this project. He offered invaluable critiques and motivation that greatly informed the shape of this project. I would like to express my sincere gratitude to all my co-workers and supportive buddies for providing support and guidance in different phases of this endeavor.
The research was made more meaningful by your involvement, contributions, and discussions. Finally, I thank my family for their constant support, comprehension and tolerance throughout this educational experience. They gave the inspiration that helped them pass through the difficulties and move on in their studies. The contribution of each of the individuals named in this section was extremely significant regarding the accomplishment of my dissertation and I would really appreciate it.
i ABSTRACT Technologies are becoming increasingly complicated and increasingly interconnected, the world trend is gradually shifting to automation. Hence, the thesis focuses on the study of a sign recognition system and an impact on control parameters settings. Speed limit signs are detected and classified using state-of-art computer vision methods based on images or videos obtained from moving vehicle cameras. As soon as the system recognizes information from the speed limit signs, it automatically makes changes in different car features such as speed, breaking system, and transmission for compliance to the traffic law.
Combining machine learning algorithms to improve performance as well as ensure safe driving by the systems through the deep learning models to participate in traffic. Besides, the outcome presents strong grounds for further theoretical and practical studies directed on construction of additional advanced autonomous guidance tools. In addition, it enhances development of smart, secure applications in automotive as well as transport sector. ii TÓM TẮT Công nghệ ngày càng phức tạp và ngày càng kết nối với nhau, xu hướng thế giới đang dần chuyển dịch sang tự động hóa.
Do đó, luận án tập trung nghiên cứu hệ thống nhận dạng dấu hiệu và ảnh hưởng đến việc cài đặt các thông số điều khiển. Biển báo giới hạn tốc độ được phát hiện và phân loại bằng phương pháp thị giác máy tính tiên tiến dựa trên hình ảnh hoặc video thu được từ camera của phương tiện đang di chuyển. Ngay khi hệ thống nhận biết thông tin từ các biển báo giới hạn tốc độ, nó sẽ tự động thực hiện các thay đổi ở các tính năng khác nhau của ô tô như tốc độ, hệ thống phanh, hộp số để tuân thủ luật giao thông. Kết hợp các thuật toán machine learning để cải thiện hiệu suất cũng như đảm bảo việc lái xe an toàn của hệ thống thông qua các mô hình deep learning để tham gia giao thông.
Ngoài ra, kết quả còn đưa ra cơ sở vững chắc cho các nghiên cứu lý thuyết và thực tiễn sâu hơn nhằm xây dựng các công cụ hướng dẫn tự động tiên tiến bổ sung. Ngoài ra, nó còn tăng cường phát triển các ứng dụng thông minh, an toàn trong lĩnh vực ô tô cũng như giao thông. iii CONTENTS ACKNOWLEDGEMENT. iv LIST OF TABLES.
vi LIST OF FIGURES AND CHARTS. vii LIST OF ABBREVIATIONS. viii CHAPTER 1: INTRODUCTION .1 Theoretical foundations of detection algorithm .3 Fully Connected Layers .5 Convolutional Filters and Feature Learning .2 Theoretical foundations of recognization algorithm .2 Multi-Scale Analysis .6 Non-Maximum Suppression (NMS) .3 Theoretical foundations of Mechanicals .4 Theoretical foundations of programming .1 Python programming language. 16 CHAPTER 3: DESIGN AND IMPLEMENTATION .1 Mechanical design selection analysis .2 Part list of mechanical for one block .2 Class and functions in ESP8266, Arduino Nano .3 Model training results.
59 CHAPTER 4: RESULT EVALUATION .2 Electrical system, signal inspection .3 Control application inspection .5 Location testing algorithm inspection. 69 CHAPTER 5: CONCLUSION AND RECOMMMENDATIONS .2 Limitation and future work. i vi LIST OF TABLES Table Table name Page Table 2.3 Raspberry Pi 4 Specifications 10 Table 2.3 Types of Camera 12 Table 2.4 Python data types 16 Table 2.4 Logical operators in Python 16 Table 3.1 Overall parameters 18 Table 3.1 Table of parts 20 Table 3.2 DC-DC buck converter MP2307 specifications 25 Table 3. OV5647 specifications [5] 27 Table 3.2 Arduino Nano specifications [7] 29 Table 3.2 I2C UART SPI TTL 8 chanel specifications 31 Table 3.4 Bill of materials table in Vietnamese currency 48 Table 3.4 Traffic Sign Types 49 Table 4.4 Testing result 68 vi LIST OF FIGURES Figure Figure name Page Figure 2.1 Theoretical foundations of recognition algorithm 2 Figure 2.2 Theoretical foundations of detection algorithm 6 Figure 2.3 Theoretical foundations of electrical and electronics 9 Figure 2.4 Theoretical foundations of programming 15 Figure 3.1 Mechanical design 18 Figure 3.2 Electrical design 25 Figure 3.3 Control design 35 Figure 3.
Operation inspection 64 vii LIST OF ABBREVIATIONS IoT: Internet of Things RPI: Raspberry Pi LED: Light-emitting diode PIC: Peripheral Interface Controller HTML: HyperText Markup Language CSS: Cascading Style Sheets I2C: Inter-Integrated Circuit SMD: Surface Mount Device RAM: Random Access Memory viii CHAPTER 1: INTRODUCTION 1.1 Problem Effective speed governance in modern transport systems forms part of road safety measures and traffic control. Exceeding the speed limit contributes greatly to traffic accident and thus requires novel ways of enforcing the same speed limit. These traditional systems of velocity control heavily depend upon motorists’ vigilance thus it is prone to human faults, tiredness, or distraction. Considering how essential this problem is, this study examines the creation of an automatic identification of velocity limitation indications device and discusses it’s influence for adapting automobile parameters.2 Objectives This research aims at developing an effective detection scheme, able to accurately identify speed limit signs.
This system seeks to use computer vision technologies that help spot speed restriction details from a digital video, image or screen grabbed photo. Furthermore, the study aims at establishing whether the data collected can be used to change some vehicle characteristics like speed, brake intensity, and transmissions. Ultimately, this objective will involve integrating smart speed limit enforcement systems into cars so as to improve safety on roads and enhance traffic flow.3 Solution This will entail incorporating contemporary computer vision algorithms as well as advanced machine learning approaches for enhanced and immediate determination of speed limit indicators for optimal outcomes. It will use a different set of speed limit image data for training and validating purposes so that it can be applicable in different environments.
The system works through recognizing information on limits of speed and communicates with a vehicle’s management system in order to make corresponding adjustments, including modification of parameters thus ensuring observation of speed set up rules. Such a holistic approach is intended to develop a smart and reactive model which does not just detect the speed limits, but also improves the safe and effective driving pattern. In the following chapters, the discussion will encompass the methods used in this study, the experimental findings, and how it has provided solutions that improve the efficiency of vehicle parameters.1 Theoretical foundations of recognition algorithm Convolutional Neural Networks (CNNs) are a class of deep learning models specifically designed for tasks involving visual data, such as image recognition and classification. The theoretical foundations of CNN recognition algorithms include key concepts and operations that enable the network to effectively learn and extract features from images.1 Convolutional Layers Convolution Operation: CNNs use convolutional layers to apply convolution operations to input images.
This operation involves sliding a convolutional kernel (filter) over the input image to perform element-wise multiplications and summations, producing feature maps that highlight certain patterns or features. Convolution Operation 2 Pooling Layers: Pooling layers, typically max pooling, are used to down sample the spatial dimensions of the feature maps, reducing the computational complexity and retaining the most salient information.2 Activation Functions Rectified Linear Unit (ReLU): The ReLU activation function introduces non- linearity to the model by replacing negative values with zero. This helps CNNs learn complex patterns and relationships within the data.3 Fully Connected Layers After multiple convolutional and pooling layers, the network often includes one or more fully connected layers. These layers connect every neuron to every neuron in the previous and subsequent layers, enabling the network to learn high -level representations.
Full Connected Layers 2.4 Backpropagation CNNs employ backpropagation for training. During the training process, the network adjusts its weights and biases based on the computed error, minimizing the difference between predicted and actual outputs.5 Convolutional Filters and Feature Learning The convolutional filters in CNNs act as feature detectors. Through the training process, these filters learn to recognize various low to high-level features in the input data, such as edges, textures, and object parts. Convolutional Filters and Feature Learning 2.6 Transfer Learning Transfer learning is a technique where pre-trained CNN models on large datasets (e., ImageNet) are fine-tuned for specific tasks.
This leverages the learned features from the broader dataset, improving performance on smaller, task -specific datasets.7 Batch Normalization Batch normalization is applied to normalize the input of each layer, enhancing training stability and accelerating convergence. It normalizes the mean and variance of the input within a batch.