Vehicle License Plate Detection Using Deep Learning
Omar Gheni Abdulateef, Abdulqadir Ismail Abdullah, Saadaldeen Rashid Ahmed, Mahdi Salah Mahdi
Abstract
This study focuses on using a Deep Convolutional Network trained with data from license plates to automatically categorize and geolocate vehicles (DCNN). Toll collection, accident reconstruction, and the identification of suspicious vehicles are just some real-world applications that use license plate readers. The study recommended using a vehicle classifier based on deep learning to pinpoint the location of license plates and license numbers simultaneously. Bounding quadrilaterals are provided by the classifier instead of bounding rectangles, which provides a more accurate indication for vehicle registration estimation to license plate localization. This task was accomplished using the Python programming language and various deep learning libraries. Since the training of the proposed DCNN model began with a weight that had already undergone a certain number of iterations in a model without a classification head, the total number of training iterations will be close to 10,000 when taking into account the transfer learning component of DCNN. Because of transfer learning, the DCNN model could begin at a good place, making it simpler to enhance functional heads at once. According to the study's characterization of the task at hand-vehicle number estimation as well as license plate segmentation and vehicle-the DCNN achieved 98.8% accuracy in classification.