Deep Learning–Driven Automatic License Plate Recognition: An Extensive Review of Architectures and Applications

The Automatic License Plate Recognition (ALPR) is now an important element of intelligent transportation systems, surveillance systems, toll collection, smart parking, and law-enforcement usage. In the last twenty years, ALPR has developed to be the classical image-processing pipelines, which are simplified to edge detection, thresholding, and morphological operations, to the more modern deep-learning-based systems, with CNNs, segmentation-free networks, object detectors like YOLO/SSD, and transformer-based recognition systems. The survey summarizes the knowledge of more than twenty high-impact ALPR studies (including classical highly referenced ones as well as recent deep learning (DL) models). Besides that, we cite the cross-domain medical-image article of Optic Disk Extraction and Hard Exudate Identification in Fundus Images using Computer Vision and Machine Learning as a source of underlying CV methods like thresholding, morphological filtering and adaptive segmentation. The survey overviews the development of ALPR, general pipelines of processing, issues related to the processing, dataset comparison, as well as the new tendencies including self-supervision, transformer-based ALPR, and edge-enabled real-time implementation.