Perspective Vehicle License Plate Transformation using Deep Neural Network on Genesis of CPNet

Abstract

Recent development in vehicular industries and increased number of cars in modern society leads the people to pay more attention on Vehicle License Plate Recognition (V-LPR). V-LPR plays a major role in traffic related application such as road traffic monitoring, vehicle parking lots access control etc. Existing state of the art V-LPR systems in real world deployment works under restricted conditions, such as static illumination, fixed background etc. Most of them fails to work when any of the above given conditions are violated. Hence to address this issue, a novel V-LPR system is designed using modern deep learning framework called "Capsule Network". The proposed system is robust and works fine in any condition. Further, the proposed method aims to improve the processing time by integrating the segmentation process within the CN framework which involves the training and recognizing of entire license plate cropped region. Moreover, the feature extraction is performed by CN framework over a segmented alphanumeric character. Finally, Data augmentation technique is also used as a supplement to the CN framework to strengthen the process of training with various orientations like rotation, shift and flip for improving the recognition task.