AUTOMATIC NUMBER PLATE RECOGNITION SYSTEM USING IMAGE PROCESSING AND MACHINE LEARNING TECHNIQUES

Abstract

Automatic Number Plate Recognition (ANPR) systems is a computer vision technology that aims to automatically recognise the vehicle's identification number from the vehicle image. Therefore, it is an essential component for automating many surveillance and control systems, such as: tracing stolen vehicles, highway toll collection, and parking entrance. This thesis presents an ANPR system for UK vehicle's front number plate recognition, which aims to achieve a highly efficient and fast system. It proposed two methods for detecting and extracting the number plate. These methods are based on identifying the number plate regions either by its edges feature with a KNN classifier or its HOG feature with an SVM classifier. The character segmentation stage goes through a combination of techniques of image processing algorithms, besides, to take advantage of the prior knowledge of the UK number plate. The recognition is performed through the use of KNN classifier. In order to analyse the performance and efficiency of the developed system, a data-set contains 150 vehicles are used, in addition to a complementary data-set for personalised, multiple, inclined, and European number plate. Several experiments have been carried out using these data-sets. These experiments results have shown that the developed system is highly efficient with accuracy rate of 95 % and 98.6 % for number plate detection, 98.5 % and 89.8 % for character segmentation, 96 % and 95 % for character recognition, 92.8 % and 93 % for the ANPR system with an average exaction time of 1.74 sec and 0.16 sec for Edge-KNN and HOG-SVM respectively.