A Study of Algorithms for Handheld License Plate Recognition System

Abstract

Automatic license plate recognition has attracted many researchers’ attention in recent years, and their efforts have resulted in many installed business applications. However, most existing installations are non-moving, such as automatic parking control and payment systems. In order to provide flexibility for implementation, this article develops a set of algorithms suitable for realizing license plate recognition (LPR) capability on handheld mobile devices. The proposed LPR algorithm consists of three major steps, namely, license plate location, character segmentation, and character recognition. Plate location is achieved mainly by extracting image outlines of probable areas. For pictures of acceptable quality, two-dimensionally reduced images are used to save running time. To separate the characters with higher accuracy, the character segmentation step applies the techniques of hybrid rotation correction. Finally, a one-layer artificial neural network is deployed for character recognition. We have implemented the proposed LPR algorithm on a DSP-based development board. The experimental results show that, the total average rate of successful recognition is 87.2% within two seconds. This is based on a sample space of 539 pictures including the ones containing more than a single license plate, taken with various shooting angles and under a range of weather and light conditions.