A HYBRIDIZED APPROACH TO EFFICIENT AND ROBUST LICENSE PLATE DETECTION AMD RECOGNITION

Abstract

With the exponential increase in the number of private and public transportation systems, automatic detection and recognition of car license plates has translated into a pivotal technology that has enabled e ective management of modern cumbersome terrestrial tra c. The technology has several diverse applications such as tra c monitoring, law enforcement, real time intelligence gathering, border management, intelligent transportation systems etc. At the same time there are many typical and challenging problems linked to it such as the variety of plate formats, varying geometry, non-uniformity in illumination, diverse weather conditions, complexity of backgrounds etc. The research brought out in this dissertation puts forward a novel method to counter yet another challenge for implementation of license plate recognition for real time tra c typically in an expressway environment, encompassing critical issues such as computation complexity and computation time, without compromising accuracy. Focus has been given on the enhancement of those sub-modules of License Plate Detection systems which are primary contributors towards processing time and computational complexity. A novel idea of a hybridized approach to counter the issues of cross-platform integration and exibility has been put forward, wherein the accuracy of algorithm proposed will not be dependent on the resolution of the input image. To further e ectively leverage the results of successful detection of the license plate, a novel technique of character segmentation for accurate and e cient extraction of license plate number from the detected license plate has also been proposed. The proposed character segmentation technique is designed keeping in focus the issues related to adaptability to variable plate backgrounds and illumination without compromising the factors of processing time and computation complexity.