交通號誌辨識在智慧型駕駛中扮演著很重要的角色，能夠即時提醒，並提高行車時的行車安全。本篇論文提出基於類神經網路的學習方式，利用regional occlusion mask的方法，達到改善對於遮蔽交通號誌因遮蔽的關係而失去號誌整體特徵，導致辨識錯誤的問題，並且提高辨識時的準確度。本篇系統也對遮蔽號誌做全域性和區域性的特徵學習，遮蔽號誌透過全域性的特徵學習後，再經由對遮蔽號誌五個區域做區域性的學習，目的是希望能夠解決部分遮蔽的影像所帶來失去整體特徵的問題，我們也利用自己收集的號誌影像(以台灣的北市、新北市)區域為主、德國公開的資料集GTSRB來做實驗。

Traffic sign recognition is very important in the intelligent driving. It enables to remind drivers in real time, and enhance the driving safety. In this paper, we propose a method based on the neural network learning, utilizing the regional occlusion mask method, to improve occlusion traffic sign recognition. This paper utilizes global and local to learn feature of occlusion traffic sign, after learning global feature of occlusion traffic sign, then learning for five local regions of occlusion traffic sign. It expect to solve partial  occlusion image, to lose the whole feature problem. We also collect traffic sign images from Taipei City and New Taipei City of Taiwan. The proposed system is tested by our own dataset and German public dataset. The experimental results are promising.

目錄
目錄	III
圖目錄	V
表目錄	VI
第一章 研究動機與目的	1
第二章 相關研究	2
第三章	類神經網路基本介紹	5
3.1	卷積層(Convolution Layer)	6
3.2	最大池化層(Max Pooling Layer)	7
3.3	線性整流函數(Rectified Linear Unit)	8
3.4	全連接層(Fully Connected Layer)	8
3.5	Softmax函數	9
3.6	Loss Function 損失函數	9
3.7	反向傳播(Backpropagation)	12
第四章	研究方法	14
4.1	樣本收集	14
4.1.1 雙北市交通號誌資料庫的建立	15
4.1.2	資料擴增	16
4.2	系統架構	17
4.2.1	Global feature extraction	18
4.2.2	Local feature extraction	19
4.2.3	Loss function	20
4.3	特徵圖	21
第五章	實驗結果與分析	23
5.1	實驗設備	23
5.2 基於本篇論文系統架構與自己收	24
5.2.1	Confusion Matrix-類別表	25
5.3	GTSRB Dataset實驗	27
5.4	基於本篇架構-GTSRB Dataset批次訓練20類後的準確度	28
5.4.1	GTSRB Dataset-取五個相似的類別做微調	29
第六章 結論與未來展望	30
參考文獻…………………………………………………………………………………… 31
附錄：英文論文	33


圖目錄
圖1 偵測與辨識流程圖	2
圖2 卷積層範例	6
圖3 最大池化層範例	7
圖4 線性整流函數範例	8
圖5 全連接層範例	9
圖6 二元分類Cross Entropy損失函數	10
圖7	二元分類上Cross Entropy與Shrinkage Loss的比較圖	11
圖8 反向傳播範例	13
圖9 樣本收集	14
圖10 ㄧ些訓練影像範例	15
圖11 ㄧ些測試影像範例	15
圖 12 資料擴增影像範例	16
圖13 系統架構	17
圖14  Global feature extraction 的架構	18
圖15  Local feature extraction 的架構.	19
圖16 系統Loss Function位置	20
圖17 遮蔽與無遮蔽影像特徵圖比較	21
圖18 Local feature extraction遮蔽影像特徵圖	21
圖19 其他遮蔽影像Global / Local feature extraction特徵圖	22
圖20 本篇系統準確度比較	24
圖21 類別表	25
圖22 分類錯誤影像	26
圖23 GTSRB 影像範例	27
圖24 GTSRB Dataset相似類別影像範例	29
圖25 模糊號誌	30
圖26 仰角號誌	30
表目錄
表1 GTSRB Dataset於本篇系統訓練後平均準確率	28
表2 與別篇系統準確度比較	28

[1]	Xianyan Kuang, Wenbin Fu, Liu Yang, “Real-Time Detection and Recognition of Road Traffic Signs Using MSER and Random Forests,” International Journal of Online and Biomedical Engineering (iJOE), Vol. 14, No. 3, 2018, pp. 34- 51.
[2]	Mohit Singh, Manish Kumar Pandey, Lakshya Malik, “Traffic Sign Detection and Recognition for Autonomous Vehicles,” International Journal of Advance Research, Ideas and Innovations In Technology (IJARIIT), Vol. 4, No. 2, 2018, pp. 1666- 1670.
[3]	Abrarul Haque Siddique, Dr. Preetam Suman & Kavita Agarwal, “Traffic Sign Detection and Recognition Using Digital Image Processing,” International 
Journal of Advanced Research in Computer Science, Vol. 9, No.2, 2018, pp.204-208.
[4]	Yi Yang, Hengliang Luo, Huarong Xu, and FuchaoWu, “Towards Real-Time 
	Traffic Sign Detection and Classification,” IEEE Transactions on Intelligent 
Transportation Systems, Vol. 17, No. 7, 2016, pp. 2022- 2031
[5]	Yixin Chen, Yi Xie and Yulin Wang, “ Detection and Recognition of Traffic
Signs Based on HSV Vision Model and Shape Features,” Journal of Computers, Vol. 8, No. 5, 2013, pp. 1366- 1370.
[6]	Markus Mathias, Radu Timofte, Rodrigo Benenson, and Luc Van Gool, “ Traffic
	Sign Recognition - How Far Are We From The Solution?” International Joint 
	Conference on Neural Networks (IJCNN), 2013.
[7]	Ya-Li Hou, Xiaoli Hao, and Houjin Chen, “ A Cognitively Motivated Method for Classification of Occluded Traffic Signs, “ IEEE Transactions On Systems, Vol. 47, No. 2, 2017, pp.2168-2216.
[8]	Dip Nandi , A.F.M. Saifuddin Saif , Prottoy Paul, Kazi Md. Zubair , Seemanta 
	Ahmed Shubho , “ Traffic Sign Detection Based on Color Segmentation of 
	Obscure Image Candidates: A Comprehensive Study,” I.J. Modern Education 
and Computer Science, Vol. 10, No. 6, 2018, pp.35-46.
[9]	楊修銘 劉昆灝 劉昭麟, “Traffic Sign Detection and Recognition In Noisy 
Outdoor Scenes,” 台灣人工智慧學會第七屆人工智慧與應用研討會, 1999, pp.555-560.
[10]	Zhe Zhu, Dun Liang, Songhai Zhang, Xiaolei Huang, Baoli Li, Shimin Hu, “Traffic-Sign Detection and Classification in The Wild,” The IEEE Conference on Computer Vision and Pattern Recognition(CVPR), 2016.
[11] Jianan Li, Xiaodan Liang, Yunchao Wei, Tingfa Xu, Jiashi Feng, and Shuicheng 
Yan, “Perceptual Generative Adversarial Networks for Small Object Detection,” CVPR, 2017.
[12] Weitao Wan, Jiansheng Chen, “Occlusion Robust Face Recognition Base on 
Mask Learning,” IEEE International Conference On Image Processing(ICIP), 2017.
[13] Satoshi Iizuka, Edgar Simo-Serra, Hiroshi Ishikawa,” Globally and Locally 
Consistent Image Completion,” ACM Transactions On Graphics, Vol. 36, No. 4, 2017, pp.14.
[14] Deepak Pathak, Philipp Kr¨ahenb¨uhl, Jeff Donahue, Trevor Darrell, Alexei A. 
Efros, “Context Encoders: Feature Learning by Inpainting,” IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016.
[15] Rui Min, Abdenour Hadid, Jean-Luc Dugelay, “Efficient Detection of Occlusion 
Prior to Robust Face Recognition,” The Scientific World Journal, 2014.
[16] Jie Zhang, Meina Kan, Shiguang Shan, Xilin Chen, “Occlusion-Free Face 
Alignment: Deep Regression Networks Coupled with De-corrupt AutoEncoders,” IEEE, 2016.
[17]	Dan Ciresan, Ueli Meier, Jonathan Masci and Jurgen Schmidhuber,”A Committee of Neural Networks for Traffic Sign Classification” IEEE, 2011.
[18] Longchao Yang, Peilin Jiang, Fei Wang, Xuan Wang ,“Region Based Fully 
Convolutional Siamese Networks for Robust Real-Time Visual Tracking,” IEEE International Conference on Image Processing (ICIP), 2017.
[19] Xiankai Lu, Chao Ma, Bingbing Ni, Xiaokang Yang, Ian Reid and Ming-Hsuan 
Yang, “Deep Regression Tracking with Shrinkage Loss,” ECCV, 2018.