ETHIOPIAN LICENSE PLATE CHARACTER RECOGNITION USING MACHINE LEARNING

Abstract

Intelligent Transport Systems (ITS) rely on license plate detection and recognition (LPDR) systems. LPDR systems are used in surveillance systems such as traffic monitoring, border control, highway toll booths, and parking entrance and exit management. Recent advances in artificial intelligence have sped up the process of identifying vehicles and other objects on highways. However, the LPDR system is still an unsolved problem for many researchers. Several methods have been proposed, including deep learning techniques, but these methods are only applicable to specific regions or datasets collected privately. This research aims to develop a system that automatically reads and recognizes characters from Ethiopian car license plates. This is a critical research area because manual license plate recognition systems are severely challenged by the increasing volume of traffic on roads. The proposed method uses machine learning and computer vision techniques to recognize Ethiopian license plates from digital images. To achieve this, we compared the various existing computer vision techniques used in automatic number plate recognition (ANPR) and provided a thorough understanding of the operation and mode of use of the most commonly used machine learning algorithms in ANPR. We also created a car image dataset from scratch, in addition to developing the ANPR application, which is required for both training the machine learning algorithms and evaluating the performance of the developed system. The developed system detects a car in a given image using YOLO (You Only Look Once), a real time object detection algorithm. The detected car image is then fed into Warped Planar Object Detection Network (WPOD-NET) to localize the license plate. The cropped license plate was then segmented through different image processing operations using the OpenCV library. After successfully segmenting the characters, Recognition of characters is done using two machine learning Algorithms namely KNN and SVM. The experimental results of the current study significantly improved the character recognition rate compared to a similar study done. The overall success rate of the developed system is 98.5%. This research has made significant contributions to the field of LPDR. The developed system can be used in a variety of applications, such as traffic monitoring, border control, and parking management.