Document 1 2 - Automatic License Plate Recognition using Python and Open CV PDF

Preview

Summary

Automatic License Plate Recognition using Python and Open CV...

Description

TOPIC: Development of a robust license plate recognition system using convolutional neural networks and secure identfication features.

Group Names: C118315 Sabah Abdiaziz Hassan C118757 Abdishakur Farah Mohamud C118458 Anfac Mire Salad C118331 Hassan Osman Elmi

JAMHURIYA UNIVERSITY OF SCIENCE AND TECHNOLOGY(JUST)

CHAPTER 2

2.1. Introduction AVRI or License Plate Recognition (LPR) has been one of the most useful approaches for vehicle monitoring in recent years. It can be used in a variety of public settings for a variety of objectives, including traffic safety, Car Parking System, Automated Toll-TAX Collection, and Enforcement The four stages of the AVRI algorithm are as follows: (1) Take a picture of the vehicle (2) Take a picture of the license plate detection (3) character segmentation (4) character classification (5) plate verification recognition. At the moment, number plate recognition and In many cases, the recognition processing time is less than 50 milliseconds [4]. systems. The success of the fourth phase is determined by the second. In the third stage, you'll be able to discover the vehicle's license plate. Each character should be separated. These systems use various methods to find the car number plate on the vehicle and then extract the vehicle number from the picture. The majority of AVRI systems employ common methodologies such as ANN (Artificial Neural Network), Optical Character Recognition, MATLAB, Neural Network, Machine Learning, and many other methods to detect license/number plates; however, for detecting vehicle plates, a basic methodology is used.

Car license plate image.

Image Normalization & Binarization

Character Region Extraction

Character Segmentation

Optical Character Recognition

Identified License Plate Number

2.2 Overview of license plate recognition The majority of number plate algorithms are classified into many categories based on different methodologies. In this project, I'll use EasyOCR in conjunction with a Tensorflow model to recognize license/number plate numbers in real time. Vehicle number plates may be readily identified with the aid of EasyOCR and Tensorflow models. The following criteria should be considered while detecting a vehicle number plate. (1) Plate Size: Number plates for two and three wheelers must be 200 x 100 mm in size, according to the Central Motor Vehicle Rules of 1989. Light motor vehicles must have a 340x 200 mm number plate, whereas passenger cars must have a 500 x 120 mm number plate. The number plate on medium and large commercial trucks should be 340 x 200 mm in size. (2) Plate Location: Number plates should be visible AVRI cameras from all sides of the vehicle, even in brilliant sunshine. (3) Plate background: Depending on the vehicle type, the background color of a plate might be varied. For example, the backdrop of a government vehicle number plate may differ from that of other public cars. (4) Screws or nut-bolds: A plate might have a screw, which could be a character.

2.3 Machine Learning and Computer Vision Based Algorithms

As technology continues to resemble the human brain, AI has piqued public curiosity for decade Let's look at the link among AI, machine learning, and computer vision to see where these advancements are headed. Machine learning is a subset of AI, and computer vision is also a subset of machine learning. AI is the umbrella term for these subjects. Computer vision, on the other hand, is a direct subset of AI. Machine learning and computer vision are two disciplines that have grown increasingly inextricably linked. Computer vision has advanced in terms of recognition and tracking thanks to machine learning. It provides efficient capture, image processing, and object focus methods for computer vision. As a result, computer vision has expanded the capabilities of machine learning. A digital picture or video, a sensor device, an interpreting device, and the interpretation stage are all part of the process. In the interpreting device and interpretation step of computer vision, machine learning is applied. Machine learning, on the other hand, is a bigger area, as seen by the algorithms that may be applied to various fields. The analysis of a digital recording, for example, is carried out using machine learning methods. On the other hand, computer vision is largely concerned with digital pictures and movies. Information engineering, physics, neuroscience, and signal processing are all domains where it has connections. 2.3.1 Deep learning Vs Convolutional Neural Networks Deep Learning is a subset of Machine Learning that uses Deep Neural Networks (DNNs)which are neural networks with at least three or four layers (including the input and output layers). However, some individuals (particularly non-technical people) consider any neural network to be Deep Learning, regardless of its depth. Others think a 10-layer neural network is too shallow. Convolutional Neural Networks (CNNs) are a type of neural network design that is quite popular. They're great in image processing, but they're also good at a lot of other things (such as speech recognition, natural language processing, and more). Because state-ofthe-art CNNs include a lot of layers (at least dozens), they're considered Deep Learning. However, given a basic assignment, you can design a shallow CNN, which isn't (truly) Deep Learning.

2.3.2 Supervised Learning Supervised learning is a method of developing artificial intelligence (AI) that involves training a computer system on input data that has been labeled for a certain output. When provided with never-before-seen data, the model is trained until it can discover the underlying patterns and correlations between the input data and the output labels, allowing it to produce correct labeling results.

Supervised learning excels in classification and regression issues, such as determining the category of a news item or forecasting the number of sales for a future date. The goal of supervised learning is to make meaning of data in the context of a given problem.

2.3.3 Unsupervised Learning Unsupervised learning, also called as unsupervised machine learning, analyzes and clusters unlabeled information using machine learning techniques. Without the need for human interaction, these algorithms uncover hidden patterns or data groupings. It is the best option for exploratory data analysis, cross selling techniques, consumer segmentation, and picture identification because of its capacity to detect similarities and contrasts in information.

2.4 Convolutional Neural Networks

A Convolutional Neural Network (ConvNet/CNN) is a Deep Learning system that can take an input picture, assign relevance (learnable weights and biases) to various aspects/objects in the image, and distinguish between them. When compared to other classification methods, the amount of pre-processing required by a ConvNet is significantly less. While basic approaches

need hand-engineering of filters, ConvNet can learn these filters/characteristics with enough training. The design of a ConvNet is inspired by the organization of the Visual Cortex and is akin to the connection pattern of Neurons in the Human Brain. Individual neurons can only respond to stimuli in a small area of the visual field called the Receptive Field. A number of similar fields can be stacked on top of each other to span the full visual field.

2.5. CNN transfer learning In reality, training the CNN model from scratch is rather unusual due to two factors. To begin, training a full CNN with random starting weights on large-scale datasets utilizing several GPUs takes more than two weeks. Second, the deep CNN model is made up of a large number of parameters that must be learnt using a large dataset such as ImageNet, which has 1.2 million pictures divided into 1000 classes. However, collecting and annotating such a large volume of data takes a long time. To address these issues, we use a transfer learning strategy that involves fine-tuning CNN and then employing it as a feature extractor.

2.6 Mobile application A mobile application or app is a software application designed to run on a mobile device. However, in this project we are hoping to create an app that verifies the car that needs to go pass the university/business toll gate to make sure is the right person.

2.7 Related works There have already been several research and works on automatic vehicle identification based on the detection and recognition of license plate numbers. Depending on the plate features of each country, researchers use different approaches and algorithms. To acquire important information on ANPR systems, several academic articles were read....