Bigailab-trvlp-20k: Turkish License Plate Detection Using Deep Learning with A New Benchmark Dataset

Abstract

In order to recognize the license plates, the license plate area must first be detected on the image. There are effective machine learning methods in vehicle plate detection. However, for these methods to work effectively, it is important that the images have a certain quality. Preprocessing is necessary for these methods to work effectively on images taken under any circumstances. In this study, datasets were prepared using these images for deep learning methods. A data tagging web application has been developed to create the dataset. As a result, 2 datasets were created for Turkish vehicle license plates with 20.404 images. With this dataset, a deep learning model for plate detection using the TensorFlow deep learning library is trained. The results on the test images are at the desired level. © 2019 IEEE.