ALPR - An Intelligent Approach Towards Detection and Recognition of License Plates in Uncontrolled Environments

Abstract

Most existing Automatic License Plate Recognition (ALPR) approaches focus on images containing approximately frontal views. The considerable variation of LP across complicated environments and perspectives remains a massive challenge for a robust ALPR. This work proposes a comprehensive ALPR paradigm emphasizing unrestricted express screenplays in which the LP may be significantly influenced by diverse shooting angles, illumination circumstances, and complicated surroundings. This system integrates a Spatial Transformer Network, which can catch and repair numerous distorted LPs in an image so that all the plates are consistently aligned. Then, a convolutional neural network is sketched to determine LP characters containing various font styles and sizes. We evaluated the system with a data set containing annotations for a challenging LP image set from multiple areas and acquisition states. The experimental outcomes reveal that our proposed ALPR paradigm attains adequate recognition accuracy compared to existing methods.