An Efficient Deep Learning Approach for Automatic License Plate Detection with Novel Feature Extraction

Abstract

In the domain of traffic management, road toll collection, and parking lot systems, vehicle number plate detection and identification play a pivotal role. Unlike conventional methods that treat license plate detection and character recognition as separate tasks, the system simultaneously addresses both challenges within a single neural network. Our Proposed methodology capitalizes on the efficiency and accuracy of the one-stage object detection algorithm known as YOLO (You Only Look Once) to locate license plates under diverse and challenging conditions. To augment the quality of input images with low resolution or poor clarity, we employ super-resolution generative adversarial networks (SRGANs). The image enhancement process substantially improves the visual quality of captured license plate images, facilitating more precise character recognition. Quantitative assessment of propounded system reveals compelling results. The license plate detection component achieves an outstanding average accuracy rate of 98.5%, surpassing previous methods by 15.2%. This comprehensive approach not only reduces the dependency on manual labour but also elevates processing precision. It seamlessly integrates into existing transportation infrastructure, resulting in heightened operational efficiency, reduced traffic congestion, and enhanced security measures. The rapid evolution of neural networks and deep learning techniques has streamlined the deployment of such applications, revolutionizing the field of traffic monitoring and management with unprecedented ease and precision. One-stage object detector, widely referred to as YOLO, is used to find licence plates in difficult circumstances.