基於樹莓派與深度學習實現車牌辨識系統之研究

本論文使用樹莓派搭配OpenCV與CNN研究製作辨識台灣新式與舊式兩種的車牌辨識系統，透過OpenCV進行車牌影像前處理成功定位車牌與分割字元後，使用CNN將分割好的字元執行訓練，最後將要辨識的新式與舊式車牌同樣進行前處理並運用訓練好的CNN模組做辨識，於事先連接好的顯示器就可直觀得到車牌的辨識結果。本論文所開發的車牌辨識系統的適用於靜態的車牌影像，由於體積小方便攜帶與不佔空間的特性使其容易擴充與安置於各種環境，而訓練與測試影像的辨識正確率分別為99.87%與99.99%，足以證明卷積神經網路是一個很好的學習與辨識車牌中不同字體的模型。

關鍵字

車牌辨識； CNN ； OpenCV

並列摘要

This paper uses the Raspberry Pi with OpenCV and CNN to study and identify the new and old license plate recognition systems in Taiwan. After the license plate image pre-processing is successfully performed by OpenCV, the license plate and the segmentation characters are successfully located, and the segmented characters are executed using CNN. Finally, the new and old license plates to be identified are also pre-processed and trained using the trained CNN module. The connected display can intuitively obtain the identification result of the license plate. The license plate recognition system developed in this paper is suitable for static license plate images. Due to its small size, easy to carry and space-saving features, it is easy to expand and settle in various environments, and the recognition accuracy of training and test images is 98.77% and 99.99%, respectively, which is enough to prove that convolutional neural networks are a good model for learning and recognizing different fonts in license plates