Automatic motor vehicle number plate recognition
Abstract
The purpose of this initiative is to develop automatic motor vehicle number plate
recognition (Bangla) using machine learning, identifying and taking out the numbers
of license plates from photos. By using this system we intend to help the traffic
control system in detecting any issue within a few moments. Moreover, collecting
tolls and enforcement of law can be implemented with this number plate recognition
system. Various object detection models have been used in this in various suggested
methods to identify and recognize number plates, optical character recognition and
license plate detection make up the system’s three basic building blocks. YOLOv8,
YOLOv7, YOLOv5, VGG16, RESNET50, DETR, VGG16 are the models used in
this project. Object detection models are used to detect the number plate of a vehicle
from the images. That is how the method will be able to successfully recognize
and detect the number plate. The precision, recall and mAP value of YOLOv8 is
96.4%, 84.8%, 92.9% respectively. For YOLOv7 it is 61.1%, 46%, 46.5% respectively.
For YOLOv5 it is 98.1%, 12.1%, 17.4% respectively. DETR is 6.5%, 7.5%,
8.32% respectively. For VGG16 the test accuracy is 90.14% and for ResNet50 it is
89.91%. Additionally, this system will be implemented within the web. So by using
a phone camera the car number plates would be detected with a device like a mobile
phone. To sum up, the number plate detection system has the ability to detect,
identify and be able to save the information and will help provide a reliable management
system for traffic and capturing fraud and indiscipline in the traffic control
system.