臺灣博碩士論文加值系統

先進駕駛輔助系統(ADAS)在近年來熱烈話題無人自駕車當中佔有相當重要的一部份，然而，在先進駕駛輔助系統(ADAS)眾多功能之中，交通號誌的偵測與辨識是核心功能之一，先進駕駛輔助系統(ADAS)能夠輔助駕駛員在第一時間察覺危險狀況，增加駕駛人反應時間以對緊急狀況作出裁決判斷，用以降低交通事故之發生機率，進而提升駕駛人車安全．
近年來捲積類神經網路因在影像辨識方面取得相當高的準確率，使得捲積類神經網路在近年來相當受歡迎，此論文中我們是將捲積類神經網路使用在辨識交通號誌影像資料庫GTSRB之中．
我們嘗試了許多不同架構之捲積類神經網路，希望能夠找出較高效率的架構，使得辨識率有所提升，而在於前處理方面，除了較常見之對比度前處理方法之外，我們主要使用了自適應性反雙曲線正切函數(AIHT)演算法、捲積類神經網路投票機制以及分層捲積類神經網路來提升我們的辨識率，而我們的辨識率則達到了99.52%．

Advanced driver assistance systems, or ADAS, are the foundation to help the driver in the driving process. Traffic sign detection and recognition is an important part of ADAS that can solve the concerns over road and transportation safety, ADAS assisting a driver to avoid potential safety critical situations and take an immediately decision that can reduce the number of victims of road accident.
Recently, convolutional neural networks have become extremely popular due to their unparalleled success in image recognition problems. In this thesis, we use convolutional neural networks to obtain traffic sign features from the German traffic sign recognition benchmark (GTSRB) image database.
Different CNN architectures were tested to find the best efficient for this experiment. The image preprocessing is important and has great impact on sign reorganization rate. The adaptive inverse hyperbolic tangent (AIHT) algorithm has shown the effect of gamma correction on an image especially can be used to improve image contrast. By applying the AIHT algorithm for image preprocessing, hierarchical CNN and committee CNN, the test accuracy of the traffic sign recognition experiment is 99.52%

誌謝 i
中文摘要 ii
List of Figures v
List of Tables vii
CHAPTER 1 Introduction 1
CHAPTER 2 Background 3
2.1 Neural network to deep learning 3
2.2 Convolutional neural network 3
2.2.1 Convolution layer 4
2.2.2 Max pooling layer 5
2.2.3 Fully connected layer 7
2.2.4 Faster R-CNN 8
2.3 Histogram equalization (HE) 11
2.4 Contrast limited adaptive histogram equalization (CLAHE) 15
2.5 Adaptive inverse hyperbolic tangent algorithm 16
CHAPTER 3 Experimental material 18
3.1 Introduction 18
3.2 Experimental material 18
3.3 Committee CNN 19
3.4 Use the AIHT technology for contrast enhancement the image 20
3.5 Hierarchical CNN 21
3.6 Two stage Committee CNN 22
CHAPTER 4 Experiment Results 23
4.1 Detection case 23
4.2 Recognition case 27
CHAPTER 5 Conclusion 34
Reference 35

[1]Y. LeCun, B. Boser, J. S. Denker, D.Henderson, R. E. Howard. Hubbard, L.D. Jackel, “Backpropagation applied to handwritten zip code recognition,” Neural Computation, vol. I, no. 4, pp. 541-551,1989.
[2]Abdel-Hamid O., Mohamed A., Hui Jiang, Penn G., "Applying Convolutional Neural Networks concepts to hybrid NN-HMM model for speech recognition," in Acoustics, Speech and Signal Processing (ICASSP), IEEE International Conference on, 2012.
[3]A. Krizhevsky, I. Sutskever, Geoffrey E. Hinton, "Imagenet classification with deep convolutional neural networks," in Neural Information Processing Systems, 2012.
[4]F. Schroff., D. Kalenichenko., J. Philbin, "Facenet: A unified embedding for face recognition and clustering," arXiv preprint arXiv:1503.03832, 2015.
[5]Y.Taigman, M.Yang, M., Ranzato, M. A., & Wolf, L. (2014),” Deepface: Closing the gap to human-level performance in face verification”, In: Proceedings of the IEEE conference on computer vision and pattern recognition. 2014. p. 1701-1708.
[6]Y. L., Boser, B., Denker, J. S., Howard, R. E., Habbard, W., Jackel, L. D., and Henderson, D. 1990. Handwritten digit recognition with a back-propagation network. Advances in neural information processing systems, 396–404.
[7]C.-Yi Yu, Y. C. Ouyang, T. W. Yu, “Image contrast enhancement based on a three-level adaptive inverse hyperbolic,” Journal of Chinese of Engineers Vol. 36, No. 1, January 2013, 103-113.
[8]X. Mao, S. Hijazi, R. Casas, P. Kaul, R. Kumar,and C. Rowen, “Hierarchical CNN for Traffic Sign Recognition,” IEEE Intelligent Vehicles Symposium (IV)Gothenburg, Sweden, June 19-22, 2016. pp. 541-551
[9]J. Jin, K. Fu, and C. Zhang, “Traffic sign recognition with hinge loss trained convolutional neural networks,” IEEE Trans. Intell. Transp. Syst., vol. 15, no. 5, pp. 1991–2000, Oct. 2014.
[10]S. Ren, K. He, R. Girshick, and J. Sun. “Faster R-CNN: Towards real-time object detection with region proposal networks”. In NIPS, 2015.
[11]R. Girshick. Fast R-CNN. In ICCV, 2015
[12]D. C., Meier, U., Masci, J., and Schmidhuber, J. (2012b). “Multi-column deep neural network for traffic sign classification. Neural Networks”, 32:333–338.
[13]D. C. Ciresan, U. Meier, L. M. Gambardella, and J. Schmidhuber. “Convolutional neural network committees for handwritten character classification”. In International Conference on Document Analysis and Recognition, pages 1250–1254, 2011.
[14]Simon Haykin, Neural Network A Comprehensive Foundation, Pearson Education,1994.
[15]R. C. Gonzalez, R. E. Woods, Digital Image Processing, Prentice Hall, 2002
[16]R. Gonzalez, R. Woods, S. Eddins, “Digital image processing using MATLAB,” Pearson Prentice Hall, 2004.
[17]MATLAB help, version R2016a.