Vehicle License Plate Detection and Recognition for Parking Management System

Abstract

This research presents a process for developing a license plate detection and reading system based on image processing. An artificial neural network system, in conjunction with an algorithm for image processing, is used to develop the system. The proposed system is divided into two parts. The first part involves the development of a model for detecting license plates. A number of models were considered, and their mean average precision (mAP) values were compared at a threshold of 0.5-0.95. The models selected for comparison were YOLOv5n, YOLOv5s, and YOLOv5m. These models were trained on a dataset of 601 license plate images of cars, and the resulting mAP values were 0.647, 0.665, and 0.685, respectively. The researcher chose YOLOv5s due to its similar mAP values to the YOLOv5m model and its 2.8 times fewer parameters. The second part of the proposed system involves the use of Tesseract OCR, a tool for reading data from images to characters, to read the characters from the license plates. The average percentage of reading accuracy of all letters was 78.34% and the combined accuracy of reading province names of all provinces was 79.35%.