| 摘要: | 車牌辨識系統的應用相當廣泛，例如，電子停車位管理系統、交通違規偵測系統以及被盜車輛系統。大多數的解決方法為使用典型的車牌辨識演算法，透過影像分析技術來處理，主要為三個階段，包括車牌偵測、字元切割，以及字元辨識。這些方法發展了許多年，並且不斷的改進與優化其辨識率。但是都必須著重在兩個前提情況下：一、車牌必須清晰，且不能存在汙損，光源必須均勻；二、車牌不能過於傾斜，使的拍攝角度往往需要固定其位置，否則在字元分割上將會受到影響，造成辨識不易。再者，目前的車牌辨識都採用GPU(Graphics Processing Unit)運算與高階的硬體設備，使成本過於昂貴。為了解決上述三點問題，本研究使用嵌入式系統以及採用一模型"Tiny YOLOv3"(You Only Look Once)，該模型是一種機器學習(machine learning)，基於深度學習(deep learning)的卷積神經網路(convolutional neural network)，利用卷積層(convolution layer)來擷取目標物的特徵，進而達到物件偵測效果。整個辨識過程使用兩組神經網路，第一組從影像中偵測車牌，第二組從偵測到的車牌進行影像處理並字元分割，將分割到的字元送進Tesseract-OCR進行字元辨識。實驗結果顯示，本研究所提出的方法，無須全部滿足上述兩點情況以及使用高階的硬體設備，也能夠將車牌及其字元成功辨識。;License plate recognition systems are widely used, such as electronic parking management systems, traffic violation monitoring systems, and stolen vehicle systems. Most of the solutions are used typical license plate recognition algorithms, which are processed through image analysis techniques, which are mainly in three stages, including license plate localization, character segmentation, and character recognition. These methods have been developed for many years, and their recognition rate has been continuously improved and optimized. But all of them have two important prerequisites. First, the license plate must be clear, and there must be no fouling, and the light source must be uniform. Second, the license plate can’t be too skewed so that the angle and position of shooting view are usually the same. Otherwise the license plate detection and character segmentation will be seriously affected and then cause recognition difficulty. Furthermore, current license plate recognition uses GPU (Graphics Processing Unit) operation and high-end hardware equipment, making the cost too expensive. In order to solve these three points, our research use a model “Tiny YOLOv3” (You Only Look Once). This model is a convolutional neural network based on the deep learning in machine learning. It uses convolutional layers to get the features of object and then achieve the effect of recognition. We use a total of two models in the research. Detecting license plate from the image in the first model, and then use performs image processing and character segmentation from the detected license plates, and sends the segmented characters to Tesseract OCR to do character recognition in the second model. The results show that our research can successfully recognize the license plate and its characters without fully satisfying the above two points and using high-end hardware equipment. |
