الفهرس 
A deep learning technique for vehicle license plate recognition
AbstractOnly 14 pages are availabe for public view
 |  | from | 181 |  |  |
| | | from | 181 | | |
Abstract
Automatic license plate recognition (ALPR) has been a frequent topic of research. However, most of the current ALPR research focuses on recognizing the English license plate with very little research on the Arabic license plate. In this thesis, we take a first step in training a deep learning-based model for Arabic license plate recognition. ALPR is one of the most important applications in video surveillance and computer vision. License plate (LP) recognition system commonly combines two sub-systems: LP detection which aims to locate the LP; and LP recognition which aims to recognize the characters in the LP. The conventional algorithms used in ALPR, for detection and recognition, are susceptible to multiple challenging conditions, such as variations in lighting, viewing angle or camera rotation as well as occlusion. For automatic Arabic license plates recognition (AALPR), most of the researchers in Egypt and Arabic countries have focused on recognizing characters in LP directly using many algorithms such as Sobel edge detection, histogram equalization or template matching assuming that characters have single-font, not rotated and fixed-size properties. However, with the massive power of deep learning (DL) that can greatly improve AALPR in terms of LP detection speed and LP characters recognition accuracy, even though the existence of more challenging conditions and without prior assumptions. This thesis presents a robust and efficient AALPR system based on the state-of-the-art You only look once (YOLO) object detector that is the most robust DL-based framework under different conditions (e.g., variations in camera, lighting, and background). For character segmentation and recognition, our system uses a convolutional neural network (CNN) to detect, segment and recognize characters within detected Arabic LP. This network used the Tiny-YOLOv3 architecture, as it is significantly faster and have slightly higher accuracy compared to the other reviewed architectures