Car plate detection using machine learning techniques

Abstract

With technological advancement, the use of Automatic Number Plate Recognition (ANPR) system to detect vehicle license plate has increased. The ANPR system makes use of object detection and text recognition to achieve this aim.

In a typical ANPR system, the license plate number was first captured. Then, the license plate characters were partitioned into individual characters. The last step was to read the segmented characters. When the license plate number was captured, the quality of the image may get affected by environmental factors such as illumination or raining.

This project focuses on the possible algorithm used for object detection and character recognition. For object detection, two methods were employed. For the first approach, OpenCV was directly employed on an arbitrary input image, and from there the object, which is a vehicle, and its corresponding license plate number was identified. For the second approach, the faster R-CNN approach was first employed to detect the presence of vehicles according to a certain threshold, and then OpenCV was used to identify the license plate number of the vehicle. At the same time, various environmental conditions, such as during night-time (dimmed illumination) or at sharp angles, were considered as these conditions can affect the quality of the image detected.