A Modified Viola-Jones Detector for Low-Cost Localization of Car Plates

Abstract

Over the last decade there has been a large increase of the worldwide car fleet, which implies on a larger volume of vehicles in situations that require the intervention of humans or efficient computer systems, such as in traffic surveilance and control, stolen car detection and access control to restricted areas. Those are usually dealt with by using Automatized License Plate Recognition (ALPR) based systems. This technology is used to identify vehicles on images and video, usually by identifying the license plate number. In general, ALPR based systems are composed by three sequential stages: license plate location, character segmentation and character recognition. There exists a great number of methods developed of each of those stages, using well known digital image processing and machine learning algorithms. On this study we use the Viola-Jones cascade detector along with a pre-processing step in order to perform the license plate location step on the two current models of the Brazilian license plates. The results of several detector configurations are compared and discussed. The results point to an efficient and accurate detector.