ABSTRACT | 第6页 |
摘要 | 第7-11页 |
List of abbreviation/symbols | 第11-13页 |
Chapter 1 Introduction | 第13-18页 |
1.1 Background | 第13-15页 |
1.2 Chosen Plate Format and Used Techniques | 第15-16页 |
1.3 Motivation, contribution, and organization of thesis | 第16-18页 |
Chapter 2 Literature Review and Key Technologies | 第18-27页 |
2.1 Literature Review | 第18-20页 |
2.2 Key technologies used in our research work | 第20-23页 |
2.2.1 Introduction to Artificial Intelligence (AI) | 第20-21页 |
2.2.2 Introduction to Optical Character Recognition (OCR) | 第21页 |
2.2.3 Introduction to Machine Learning (ML) | 第21-22页 |
2.2.4 Introduction to Deep Learning (DL) | 第22-23页 |
2.2.5 Introduction to Convolutional Neural Networks (CNNs) | 第23页 |
2.3 Software Programs and Tools Used in research work | 第23-27页 |
2.3.1 Python Programming Language | 第24页 |
2.3.2 Tensorflow Deep Learning Library | 第24-25页 |
2.3.3 OpenCV Computer Vision Library | 第25页 |
2.3.4 MATLAB Software Environment | 第25-26页 |
2.3.5 C++ Programming Language | 第26-27页 |
Chapter 3 License Plate Recognition System Architectures | 第27-36页 |
3.1 License plate recognition (LPR) system | 第27-28页 |
3.2 LPR systems with Three Components Architecture (3CA) | 第28-31页 |
3.2.1 Preprocessing component | 第28-29页 |
3.2.2 Segmentation component | 第29-30页 |
3.2.3 Character Recognition component | 第30页 |
3.2.4 Advantages of LPR-3CA system | 第30-31页 |
3.2.5 Disadvantages of LPR-3CA system | 第31页 |
3.3 LPR systems with Single Components Architecture (SCA) | 第31-33页 |
3.3.1 Preprocessing component (Optional) | 第31-32页 |
3.3.2 Plate Recognition component | 第32页 |
3.3.3 Advantages of LPR-SCA systems | 第32页 |
3.3.4 Disadvantages of LPR-SCA systems | 第32-33页 |
3.4 Drawbacks of traditional DIP algorithms used in LPR systems | 第33-35页 |
3.5 Advantages of machine learning techniques | 第35-36页 |
Chapter 4 Proposed LPR System Components | 第36-47页 |
4.1 Deep learning and CNNs | 第36-37页 |
4.2 Preparing license plate images for segmentation | 第37页 |
4.3 Segmentation CNN | 第37-42页 |
4.3.1 Description of segmentation CNN operation concept | 第37-38页 |
4.3.2 Segmentation CNN structure | 第38-40页 |
4.3.3 Training of segmentation CNN | 第40-41页 |
4.3.4 Advantages of segmentation method | 第41页 |
4.3.5 Disadvantages of segmentation method | 第41-42页 |
4.4 Character recognition CNN | 第42-44页 |
4.4.1 Preparations before character recognition | 第42页 |
4.4.2 Structure of character recognition CNN | 第42页 |
4.4.3 Training of character recognition CNN | 第42-43页 |
4.4.4 Advantages of character recognition method | 第43-44页 |
4.4.5 Disadvantages of character recognition method | 第44页 |
4.5 Structure of whole LPR system | 第44-47页 |
4.5.1 Advantages of proposed LPR system | 第45-46页 |
4.5.2 Disadvantages of whole system | 第46-47页 |
Chapter 5 Experimental Results | 第47-61页 |
5.1 Segmentation CNN dataset details | 第47-48页 |
5.2 Character recognition CNN dataset details | 第48-53页 |
5.3 Segmentation CNN experimental results | 第53-55页 |
5.3.1 Evaluating accuracy of the segmentation CNN | 第53页 |
5.3.2 Achieved accuracy | 第53页 |
5.3.3 Performance and size | 第53-55页 |
5.4 Character recognition CNN experimental results | 第55-56页 |
5.4.1 Achieved accuracy | 第55-56页 |
5.4.2 Performance and size | 第56页 |
5.5 Whole LPR system experimental results | 第56-57页 |
5.5.1 Accuracy achieved | 第56页 |
5.5.2 Performance and size | 第56-57页 |
5.6 Comparison results | 第57-59页 |
5.6.1 Training of SCO-CNN model | 第57页 |
5.6.2 Accuracy of SCO-CNN model | 第57页 |
5.6.3 Performance and size of SCO-CNN | 第57-59页 |
5.7 Notes about preprocessing step | 第59页 |
5.8 Analyzing errors reasons in our LPR system | 第59-61页 |
5.8.1 Lack (shortage) of Chinese characters samples | 第59-60页 |
5.8.2 Errors in license plate detection systems | 第60页 |
5.8.3 Lack (shortage) of dataset | 第60-61页 |
Chapter 6 Conclusion and Future Work | 第61-62页 |
6.1 Conclusion | 第61页 |
6.2 Future work | 第61-62页 |
References | 第62-66页 |
Appendix 1: List of Chinese Characters of License Plates and Corresponding Provinces | 第66-67页 |