基於深度學習之車牌辨識

Vehicle License Plate Recognition with Deep Learning

黃季軒(Chi-Hsuan Huang)

指導教授 ： 傅楸善

國立臺灣大學/電機資訊學院/資訊網路與多媒體研究所/碩士(2020年)

https://doi.org/10.6342/NTU202001297

摘要

本論文提出一個人工智慧(Artificial Intelligence)解決方案，用於移動載具上進行車牌辨識。於移動中的載具上裝設攝影機，其所拍攝到的車牌影像富有各式不同角度及光影變化，然而傳統電腦視覺演算法無法有效克服於不同的環境下進行車牌辨識。因此，本研究透過深度學習(Deep Learning)方式執行車牌辨識，以因應攝影機所拍攝畫面之光源、角度等環境變因，進而提升辨識準確率。此外，電源供應在移動載具上並非易事，因此電源消耗亦為重點考量，故本研究選擇輕量的網路架構。本系統之辨識流程包含兩個階段──車牌位置偵測及車牌號碼辨識。首先，透過一卷積類神經網路(Convolutional Neural Network, CNN)模型架構執行車牌位置偵測，並將偵測到的車牌影像擷取後進行旋轉校正。接著，設計另一CNN模型架構識別字元，即大寫字母(A-Z)和數字(0-9)。本研究提出之方法於白天高速行駛時可達到95.7%精確率和95%召回率。

關鍵字

車牌辨識 ； 深度學習 ； 電腦視覺 ； 人工智慧 ； 卷積類神經網路

並列摘要

In this thesis, an AI (Artificial Intelligence) solution for LPR (License Plate Recognition) on moving vehicles is proposed. The license plates in images captured with cameras on moving vehicles have unpredictable distortion and various illumination which make traditional machine vision algorithms unable to recognize the numbers correctly. Therefore, deep learning is leveraged to recognize license plate in such challenging conditions for better recognition accuracy. Additionally, lightweight neural networks are chosen since the power supply of scooter is quite limited. A two-stage method is presented to recognize license plate. First, the license plates in captured images are detected using CNN (Convolutional Neural Network) model and the rotation of the detected license plates are corrected. Subsequently, the characters are recognized as upper-case format (A-Z) and digits (0-9) with second CNN model. Experimental results show that our system achieves 95.7% precision and 95% recall at high speed during the daytime.

並列關鍵字

License Plate Recognition ； Deep Learning ； Computer Vision ； Artificial Intelligence ； Convolutional Neural Networks

參考文獻

[1] A. Bochkovskiy, C. Y. Wang, and H. Y. M. Liao, “YOLOv4: Optimal Speed and Accuracy of Object Detection,” arXiv preprint arXiv:2004.10934, https://arxiv.org/pdf/2004.10934.pdf, 2020.

[2] G. R. Gonçalves, S. P. G. da Silva, D. Menotti, and W. R. Schwartz, “Benchmark for License Plate Character Segmentation,” Journal of Electronic Imaging, Vol. 25, No. 5, pp. 1–5, 2016.

[3] A. Graves, S. Fernandez, F. Gomez, and J. Schmidhuber, “Connectionist Temporal Classification: Labelling Unsegmented Sequence Data with Recurrent Neural Networks,” Proceedings of International Conference on Machine Learning, Pittsburgh, Pennsylvania, pp. 369-376, 2006.

[4] Intel, “Intel Distribution of OpenVINO Toolkit,” https://software.intel.com/en-us/openvino-toolkit, 2020.

[5] P. Isola, J. Y. Zhu, T. Zhou, and A. A. Efros, “Image-to-Image Translation with Conditional Adversarial Networks,” Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, Hawaii, pp. 1125-1134, 2017.