Abstract
In modern life, the massive number of vehicles makes it hard for a human being to process its related information. So, it is important to build an automatic system to collect information about vehicles. The license plate is the unique identifier of a vehicle. In this paper, we propose an automatic license plate recognition system. The proposed system was based on the Faster R-CNN improved by adding an adaptive attention network for the segmentation of the license plate to retrieve the numbers and the letters of identification. Also, we add a deconvolution layer at the top of the features extraction network to detect the small size of the target license plate. To train and evaluate the proposed system, a dataset was collected for Arabic countries such as Egypt, KSA, and UAE that have similar license plates with Arabic and Indian numbers, Arabic and Latin alphabets. The dataset was collected from the internet using a python script then it was filtered and annotated manually. The evaluation of the proposed model dataset results in achieving a recall of 98.65 % and a precision of 97.46 %. The developed system was able to process images in real-time with a processing speed of 23 FPS.