AUTOMATIC LICENSE PLATE RECOGNITION (ALPR) WITH MULTIPLE OCR ENGINES

Abstract

In most of RUC (Road User Charging) free flow tolling systems, video sensors are installed on gantries in tolling points to identify vehicle registration plate number as travelers passing a tolling detection point. Through capturing images of vehicles and recognizing license plates number or other identifying characteristics automatically by OCR (Optical Character Recognition) technology, the free flow tolling system allows users to pay without stopping. High OCR accuracy rate and low error rate is very critical to ensure the right vehicle to be charged and also is a main factor to speed up ITS free flow tolling worldwide adoption. However, OCR is a tricky thing and there are many factors that affect OCR accuracy. Many popular OCR engines have limited overall OCR accuracy rate especially in vehicle license plate recognition. If free flow tolling systems only rely on one OCR engine result, the overall accuracy can not be guaranteed and many customer complaints may be received. Under this situation, in order to optimize the final recognition accuracy, the proposed solution will use multiple OCR engines and a set of strategic rules to evaluate individual OCR engine result and produce credible final OCR result. Please note that all ideas and data including analysis are from or based on IBM projects.