A Practical License Plate Recognition System on PDAs

Abstract

Vehicle license plate recognition systems have long been expected to be applied to traffic surveillance and monitoring, such as finding stolen cars, controlling access to parking lots and gathering traffic flow information. There are two common considerations in those systems: accuracy and real-time response. In our proposed electronic billing (E-Bill) system, however, we focus mainly its portability and still keep acceptable accuracy. The E-Bill system is best suited to be used in parallel parking spaces or the parking lots without access control. It provides an excellent framework for parking clerks to recognize the license plate through a PDA installed with a plug-in camera and then print out the tickets immediately. There are four major stages in the E-Bill system: capture the image, extract the license plate in the image, segment license numbers and recognize the numbers. Extraction and segmentation are elaborately processed by some well-known technologies digital image processing. Using the back propagation neural network to perform recognition is the very contribution of the system. Experimental results show that the E-Bill system can effectively recognize most of Taiwan’s license plates, which include 10 digits and 26 alphabets. The recognition rate is 90% at a low resolution of 1.3M pixels. The recognition time takes 2 seconds PDAs and less than 10-3 second on personal computers. After empirically evaluation, we can conclude that the E-Bill system is quite impressive with its excellent portability and still has competitive performance even under its inherent limitations. From a practical point of view, we also believe that the E-Bill system is worth introducing many parking areas or other application domains.