On the Application of Real-Time Deep Neural Network for Automatic License Plate Reading from Sequence of Images Targeting Edge Artificial Intelligence Architectures

Abstract

Traffic management, law enforcement, toll collection, and vehicle owner identification have become major problems in every country. Therefore, there is a need to develop the automatic license plate recognition (ALPR) system as one of the solutions to this problem. In recent years, several ALPR systems are available, those very systems are based on specific methodologies, but it is still a particularly challenging task because some of the variables such as high vehicle speed, non-uniform vehicle number plate, vehicle number language, and different lighting conditions will significantly affect the overall rate of recognition. On other hand, the expansion of video camera deployment on almost every intersection under the intelligent transportation system enables a massive size of video streams that could be analyzed if the bandwidth and computing power is available to produce real-time results. Deep neural networks (DNN) have recently been applied, as other computer vision applications in order to increase the accuracy compared to statistical and classical image processing techniques. However, the real-time response and effective accuracy yet to be assessed in addition to the use of a simple video stream localized at the traffic intersection where each camera is covering a lane of each direction is not explored. We review the performance of an ALPR system by comparing different recognition systems in term of the models and datasets and workstation and recognition performance and processing speed.