Deep Learning Based Vehicle License Plate Recognition System for Complex Conditions

Abstract

Automatic License Plate Recognition (ALPR) system that could be used as a root for several existent world Intelligent Transport System applications. For ALPR difference between jurisdictions on character dimension, spacing and therefore the existence of noise sources like heavy shades, non-uniform brightness, many optical geometries, poor distinction and so on present in license plate images makes it difficult for the recognition accuracy and scalability of ALPR system. The proposed approach offerings a strong and full proof technique with the target of precisely pinpointing vehicle name plates from critical sections in actual world. The approach is deliberate to covenant by blurred vehicle plates, changes happening climate, illumination situations as well as different traffic situations. A unique background removal method is planned to excerpt candidate areas through mainly minimizing the field to be analyzed for license plate selection. To recognize the correct license plate between the candidate areas a gushed license plate identifier based on in lines PNN consuming color saliency structures is presented. For recital assessment, a dataset containing some pictures captured from various sights under different circumstances such as road crossings, roads and express ways, daytime and night-time, numerous climate circumstances and various plate clearness. Various experimentations on the mostly applied vehicle license plate dataset and our recently added dataset proves that the planned tactic significantly overtakes modern strategies in positions of both recognition correctness and run-time efficiency.