A Study on Automatic License Plate Recognition

Abstract

Automatic license plate recognition, ALPR, is the process of detecting license plates, and recognizing the characters, present in any given image. ALPR can be applied in many situations to reduce the requirement for manned operation of various tasks. In this thesis, different methods for performing ALPR are tested and qualitatively evaluated. The main focus is on deep learning based object detectors, how they can be applied for a more modern version of ALPR and the challenges involved. A significant problem with deep learning is the amount of data needed, and this was also encountered by the author of this thesis. The data sets gathered did not have enough variation for training deep learning models with satisfactory accuracy, which was solved by generating images of license plates to increase the variation in the data sets.