License Plate Recognition System: Localization for Kenya

Abstract

This paper describes the development of a reliable and accurate License Plate Recognition (LPR) system. In view of its potential application in traffic monitoring systems and highway toll collection, LPR systems have recently attracted considerable interest as part of an Intelligent Transport System. While much commercial work has been done for Iranian, Korean, Chinese, European and US license plates little work has been done for developing country LPR systems. In general LPR consists of four stages; Image acquisition and processing, License plate extraction, License plate segmentation and License plate recognition. This paper utilizes algorithms for the extraction stage based on vertical edge detection. The segmentation stage is performed using two algorithms: division by eight and the horizontal and vertical projection profile also known as the peak to valley method. Finally two approaches of performing recognition are investigated namely template matching and artificial neural networks, particularly the multilayer perceptron. The system was implemented using Matlab 7.6 (R2008a), Microsoft Visual Studio and Wamp Server tools. The performance of the system on about seventy real images resulted in a predictive accuracy of about 86.99% using the template matching recognition algorithm after segmenting with the peak to valley method.