DEVELOPMENT OF A DEEP LEARNING BASED VEHICLE LICENSE PLATE DETECTION SCHEME
Need help with a related project topic or New topic? Send Us Your Topic
DOWNLOAD THE COMPLETE PROJECT MATERIAL
DEVELOPMENT OF A DEEP LEARNING BASED VEHICLE LICENSE PLATE DETECTION SCHEME
Chapter One:
Introduction
1.1 Background of the study
of Yuan et al. (2016) report a surge in interest in computer vision research due to advancements in intelligent transportation systems. This is due to its use in areas such as vehicle management, electronic payment systems (toll collection on motorways and parking fee payment), access control for monitoring areas with limited accessibility such as embassies, factories, military barracks, and so on, as well as identifying lost or stolen vehicles, border control, and road traffic monitoring (Du et al., 2013).
The first licenced plate detecting system was established in 1976 by the United Kingdom’s police scientific development section (Nguwi & Lim, 2015). At the time, the functions of the licence plate detection system were severely limited.
The purpose of number plate detection is to catch unlicensed drivers and auto thieves (Jenkins, 2007). In 2007, the automated licence plate recognition (ALPR) system was integrated with red-light cameras in the United States of America (USA) to capture drivers who drove through red traffic signals.
The camera captures the offender’s car plate information and processes it using automatic licence plate recognition (ALPR) (Jenkins, 2007).
The licence plate recognition system is separated into two parts: detection and recognition (Zhao et al., 2011). Detection is the capacity to locate the licence plate and generate appropriate bounding boxes that enclose the detected licence plate, whereas plate recognition tries to identify the characters represented within the bounding boxes and classify them as a licence plate.
Licence plate detection and recognition are two distinct processes, and research on these two techniques has always been conducted independently. Nguyen et al. (2015) devised and implemented different algorithms for the two procedures.
Licence plate detection is the most significant part of the licence plate recognition system since identification accuracy is determined by the detection stage (Zhao et al., 2011).
However, a licence plate recognition system is necessary for real-time applications, therefore a high detection rate is critical in order to meet the criteria of such applications.
Although various algorithms for licence plate detection have been developed over the last two decades, some of which require sophisticated cameras to provide high-quality images, others require vehicles to pass a fixed entry gate slowly or even halt completely (Nguwi & Lim, 2015).
All of these requirements are intended to provide a sharper view of the item. Despite these efforts, detecting licence plates in an open and noisy setting remains difficult.
The situation becomes more complicated, especially when licence plate numbers are not standardised, such as when they are faded, partially obscured by dirt, or taken in diverse environmental conditions (Fomani & Shahbahrami 2017).
Detecting these plates using typical approaches may yield numerous false positives (Li & Shen, 2016). To solve this challenge, cutting-edge deep learning techniques were investigated.
1.2 Significance of The Research
A licence plate detection system is necessary for real-time applications such as access control, traffic management, and electronic toll collection. As a result, such real-time applications necessitate a licence plate detection scheme that detects licence plates accurately and quickly, regardless of environmental conditions
resulting in a rapid, cost-effective, and highly accurate recognition system. This can be accomplished by utilising the latest image processing and deep learning technology.
3
1.3 Statement of Problem
Detecting plate numbers from images is a tough operation because the images include a lot of noise, reducing their quality. These noises are caused by the photograph being captured under various weather and lighting conditions, such as uneven illumination and the presence of background clutter.
Conventional techniques, which rely primarily on specific morphological procedures, have been widely utilised for detecting licence plate position, but they have limits in real-time applications due to their temporal complexity. Some techniques, such as texture and color-based algorithms, are also not effective in detecting noisy or distorted images.
To effectively and accurately detect a licence plate from any image, regardless of background complexity, at least human level accuracy is required.
This can be accomplished by utilising a deep CNN architecture that employs a deep feature extraction strategy through a transfer learning approach. This is required to increase the detection rate while decreasing the number of false positive outcomes.
1.4 Aims and Objectives
The goal of this project is to create a deep learning-based licence plate detecting technique. The research aims are:
1. To construct a dataset of vehicle photos at Ahmadu Bello University (ABU), named the ABU dataset.
2. Develop a dCNN-based licence plate detection system.
3. Test the functionality of the devised approach on the ABU, Caltech, and PKU Vehicle I.
Need help with a related project topic or New topic? Send Us Your Topic