SUDANESE VEHICLES LICENSE PLATE RECOGNITION

Abstract

Vehicle license plate recognition is a computer vision method that aims to automatically recognize the vehicle’s identification number from the vehicle image. Therefore, it is an important component for automating many control and surveillance systems, such as: road traffic monitoring, private and public entrances, highway electronic toll collection, red light violation enforcement, and theft control. Although, considerable researches have been carried out for vehicle license plate detection and recognition in different countries, however, very few previous studies have been done for the Sudanese license plate detection and recognition. Furthermore, the vehicles in Sudan are currently being identified by traffic policemen manually. Thus, this thesis presents a novel system approach for Sudanese vehicle license plate recognition, which aims to improve the efficiency and the accuracy of the license plate detection and recognition process. A simple method is proposed for detecting and extracting the license plate, which is based on identifying the plate region by analyzing Sudanese plate features. The plate’s skew angle is computed to guarantee the license plate characters were accurately segmented. The character segmentation process goes through combined techniques to improve license plate image contrast, in addition to take advantage of the prior knowledge of Sudanese license plate. The recognition is performed through two different recognizers, English character recognizer (Template Matching method) and Indian digit recognizer (a novel heuristic rules based on salient features). In order to analyze the performance and efficiency of the proposed approach a dataset for Sudanese vehicles has been created. This dataset contains 375 vehicle images. Using this new dataset, a number of experiments have been carried out. Experimental results have shown that the proposed approach is efficient with accuracy rates of 98.6% for license plate detection, 99.5 % for character segmentation, and 98.4% for recognition.