VISVESVARAYA TECHNOLOGICAL UNIVERSITY

BELAGAVI-590018, KARNATAKA

TECHNICAL SEMINAR REPORT

ON

“Computer Vision based License Plate Detection for

Automated Vehicle Parking Management System”
Submitted by

GANESH KUMAR P
1CR19EC402
Under the guidance of
Mrs. Sushma B
Asst. Professor
Dept. of ECE, CMRIT
Department Of Electronics and Communication Engineering
Feb – June 2022

Department Of Electronics and Communication Engineering

CMR INSTITUTE OF TECHNOLOGY
#132, AECS LAYOUT, IT PARK ROAD, KUNDALAHALLI, BENGALURU-560037
 DEPARTMENT OF ELECTRONICS AND COMMUNICATION ENGINEERING

CERTIFICATE

This is to certify the Technical Seminar Report entitled “Computer Vision based License Plate
Detection for Automated Vehicle Parking Management System”, prepared by Mr. Ganesh
Kumar P, bearing USN 1CR19EC402, a bona fide student of CMR Institute of Technology,
Bengaluru in partial fulfillment of the requirements for the award of Bachelor of Engineering in
Electronics and Communication Engineering of the Visvesvaraya Technological University,
Belagavi-590018 during the academic year 2019-22.

This is certified that all the corrections and suggestions indicated for Internal Assessment have been
incorporated in the report deposited in the departmental library. The seminar report has been
approved as it satisfies the academic requirements prescribed for the said degree.

---------------------- -----------------------
Signature of Guide Signature of HOD
Mrs. Sushma B Mr. R Elumalai
Asst. Professor Professor & HoD
Dept. of ECE, CMRIT Dept. of ECE, CMRIT
 ACKNOWLEDGEMENT

The satisfaction that accompanies the successful completion of any task would be
incomplete without mentioning the people whose proper guidance and encouragement has served
as a beacon and crowned my efforts with success. I take an opportunity to thank all the
distinguished personalities for their enormous and precious support and encouragement
throughout the duration of this seminar.
I take this opportunity to express my sincere gratitude and respect to CMR Institute of
Technology, Bengaluru for providing me an opportunity to present my technical seminar.
I have a great pleasure in expressing my deep sense of gratitude to Dr. Sanjay Jain,
Principal, CMRIT, Bangalore, for his constant encouragement.
I would like to thank Mr. R Elumalai, HoD, Department of Electronics and
Communication Engineering, CMRIT, Bangalore, who shared his opinion and experience
through which I received the required information crucial for the seminar.
I consider it a privilege and honor to express my sincere gratitude to my guide
Mrs. Sushma B, Assistant Professor, Department of Electronics and Communication
Engineering, for the valuable guidance throughout the tenure of this review.
I also extend my thanks to the faculties of Electronics and Communication Engineering
Department who directly or indirectly encouraged me.
Finally, I would like to thank my parents and friends for all their moral support they have
given me during the completion of this work.

Ganesh Kumar P
1CR19EC402
 ABSTRACT

With proliferation of vehicles across the world, it is getting increasingly strenuous to manage
parking in several spaces viz business parks, residential complexes, shopping malls etc. An
optimum utilization of available parking spaces and minimizing the time and effort involved in
vehicle parking, an integrated and automated vehicle parking management system (VPMS) is
necessitated. License plate contains relevant information about vehicle and its detection &
recognition in real time can be utilized to develop an automated VPMS. In this paper, a solution
is proposed for live detection and recognition of a moving vehicle’s license plate number using
Computer Vision techniques. Three different models had been studied viz HAAR Cascade and
CNN1, OpenCV2 and YOLOv3 with OpenCV3 to find the best performing model. Among these,
YOLOv3 with OpenCV outperforms other models due to its ability to detect the rectangular
bounding boxes with great accuracy. The automation of License plate detection is a two-step
process which includes detection of custom object i.e., License plate using OpenCV algorithm.
The Trained model is validated and demonstrated 100% accuracy in detection of license plate
bounding boxes along with 95% accuracy in text recognition. This module can be implemented
and integrated with other add-on systems for effective usage in various sectors.

Keywords – License Plate Detection, VPMS, YOLOv3, CDMS

I
 CONTENTS

Page No.
CERTIFICATE
ACKNOWLEDGEMENT
ABSTRACT
1. INTRODUCTION 1
PROPOSED SYSTEM OVERVIEW 2
2. SOLUTION OVERVIEW
4
2.1 DATA COLLECTION
3. TECHNOLOGY USED AND ITS CHARACTERISTICS 5
3.1 OpenCV 5
3.2 HAAR Cascade 6
3.3 YOLOv3 7
3.4 CHARACTER RECOGNITION 9
3.5 PARKING SLOT MANAGEMENT 10

4. RESULT & DISCUSSION 11

5. CONCLUSION 12
6. REFERENCES 13

II
 LIST OF FIGURES

Figure No. Title Page No.

1.1 Schematic of VPMS 2

2.1 Solution overview 3

2.2 Dataset Preparation using labelling tools 4

3.1 License Plate detection using OpenCV 5

3.2 License Plate Detection using HAAR Cascade 6

3.3 Yolov3 Network Architecture 7

3.4 Comparative assessment- Yolov3 vs other models (mAP 50) 8

3.5 License Plate detection using YOLOv3 8

3.6 Extracted License Plate by detection models 9

3.7 Augmented Images 9

3.8 License Plate Text Recognition 9

3.9 Free Parking slot detection and display 10

4.1 Accuracy and timing results for used models 11

III
Computer Vision based License Plate Detection for Automated Vehicle Parking Management System

Chapter 1
INTRODUCTION
With the advent of new technologies and revolution in auto sector, there is significant
growth in number of vehicles across the world. The management and monitoring of these vehicles is
getting increasingly tiresome and their control requires tremendous efforts as well as high capital
investments. Monitoring of vehicle traffic and management of parking lots in crowded and congested
locations like shopping malls, business parks, residential complexes etc. has emerged as a major area
of research in Intelligent Transportation System (ITS). Several manual as well semi-automated
systems and approaches have been adopted worldwide to automate the VPMS. Automated Parking
management system based on real time detection and recognition of car license plate (LP) information
is a cost effective and easily implementable solution.

Though many new methods like RFID (Radio frequency identification), WSN (Wireless sensor
network), etc., have been introduced for vehicle ID recognition but the license plate recognition based
on computer vision techniques using the live videos and object images has significant advantage over
other methods. In this method the License plate information is obtained for the moving object from
video streaming frame by frame, choosing the best recognition result for a respective car, and then
storing vehicle information in system database for further processing. Also, the real time images
captured from the parking lot provides the information on availability of parking slot and its
integration with Central Data Management System (CDMS) helps in optimum usage of available
parking slots.

The VPMS using the proposed techniques ensures a hassle-free car parking with automated parking
fee collection and the relevant information is collected at the entry/exit of commercial spaces without
the vehicle stop and requirement of any additional equipment installation on the car. Object detection
is an important feature of this system and key subject of research for the proposed solution. For the
past two decades, there have been extensive research for number plate recognition based on object
image. Initial days plate recognition systems were based on detection methods such as edge analysis,
color analysis, and the Ada-boost training. S.Z. Wang and H.M. Lee, D. Zheng et al., and P.
Rattanathammawat and T.H. Chalidabhongse [1-2] proposed an edge analysis method for license
plate detection. Ernst [3- 5] introduced a face detection method based on local structure patterns
computed by the Modified Census Transform (MCT). In the paper on “Automatic Parking
Management System and Parking Fee Collection Based on Number Plate Recognition” M. Rashid,
A Musa, M. Ataur Rahman, N. Farahana and A. Farahana [6] discussed the importance of guided
parking for effective parking space utilization. In recent years, deep learning techniques have been
applied in object detection and exhibited superior characteristics over other methods. Object detection
methods based on deep learning technique mainly includes region proposal-based methods like R-
CNN methods and those based on a unified pipeline framework like YOLO models.

The first method works by generating the region-specific details from an input image, extracts
required image features and classify the object based on convolution neural network algorithm.
Girshick, R have studied various aspects of R CNN methods in their publications for object detection
and classification. The method based on unified pipeline framework directly predicts location
information and class probabilities of objects with a single-feed forward convolution neural network
from the whole image, and does not require the generation of region proposals and post-classification.
This makes the method simple and enables the faster object detection. YOLO models are widely used
Ganesh Kumar P – 1CR19EC402 Dept. of ECE – Feb – June 2022 Page 1
Computer Vision based License Plate Detection for Automated Vehicle Parking Management System

methods and significant research is undertaken to improve and adopt these methods for real time
applications and discussed about the optimization and speed of YOLOV models. License plate
detection using YOLOv3 algorithm combined with OpenCV and Pytesseract for conversion to text
output forms the baseline model for proposed VPMS.

1.1 PROPOSED SYSTEM OVERVIEW

In this paper, embedded computer vision-based model approach is followed for vehicle parking
system. Figure1 shows the schematic for the proposed VPMS. Input to the system is a real time
video/image captured by cameras installed at entry, exit and parking areas. From the given video,
the frame with highest bounding box probability is used as the key image to process the license
plate information along with the timestamp at entry/exit of the vehicle. Information about vehicle
LP, entry/exit time and available parking slots is generated from Computer Vision based Module 1,
2 and 3 as shown in figure1. These inputs are integrated with CDMS to provide guided parking
navigation through LED display boards and to facilitate automated parking fee collection.

Fig 1.1 Schematic of VPMS

Following are the key features of proposed solution:

• Whole system is based on Computer Vision techniques providing the easy to implement and cost-
effective solution relative to traditional methods that uses ANPR cameras, occupancy sensors,
RFID card readers etc.

• Facilitates complete automation of VPMS without any human intervention.

• Easily adaptable to add on 3rd party payment gateways for parking fee collection Additionally, it
provides central data management and real time vehicle surveillance for high reliability and vehicle
security and also ensures hassle free parking management & customer satisfaction. The integrated
data can further be utilized for reporting and sentiment analysis.

Ganesh Kumar P – 1CR19EC402 Dept. of ECE – Feb – June 2022 Page 2

Computer Vision based License Plate Detection for Automated Vehicle Parking Management System

CHAPTER 2
SOLUTION OVERVIEW
The main objective of this system is real time recognition of license plates with
time stamping and identification of availability of parking slots and its integration with CDMS. The
solution overview is depicted in Figure 2.1

Fig 2.1 Solution Overview

Ganesh Kumar P – 1CR19EC402 Dept. of ECE – Feb – June 2022 Page 3

Computer Vision based License Plate Detection for Automated Vehicle Parking Management System

From Figure 2.1 we can see the Overview of the proposed system, the first part of
this system is License Plate detection and text recognition, for this the very first step is data collection
and then manually Annotation of the labels onto those data. In this paper as data, they have used Car
Images and have labelled the car license plate using open-source websites and manually exported
annotation for object detection.

After labeling the plates they have done a Comparative study on License plate Detection Models like
OpenCV, YOLOv3 and HAAR cascade model and after going through their pros and cons. They
have selected YOLOv3 as the base model for License plate detection. After detecting the license
plate, they have done Image Augmentation using OpenCV, by using this Augmented image character
Recognition is done using Pytesseract.

The text output of License plate with the timestamp is saved into the database and can be accessed
for further processes. The detailed explanation on each of the techniques used in this proposed system
are explained below.

2.1 DATA COLLECTION

Due to unavailability of labelled vehicle license plate data, 300+ images/videos had been captured to
create the training dataset. Data capture covered the broad spectrum of variations in dataset viz angle
of image capturing, illumination level, car types/color and position of license plate. The captured
images were then manually annotated using Labelling tool to define the bounding boxes for license
plate region that helped in model training for a specified IOU of 0.75

Fig 2.2 Dataset Preparation using labelling tools

For labelling dataset, we can use different labelling tools which are available in the market. Some of
the popular labelling tools are Roboflow.com and Makesense.ai. among these by using Makesense.ai
we can freely label an image and export annotation of them and proceed with the further processes.

Ganesh Kumar P – 1CR19EC402 Dept. of ECE – Feb – June 2022 Page 4

Computer Vision based License Plate Detection for Automated Vehicle Parking Management System

Chapter 3
Technology Used and Its Characteristics

LICENSE PLATE DETECTION

Three different models had been studied for license detection and evaluated for
selecting the system base model:

3.1 OpenCV
OpenCV (Open-Source Computer Vision Library) is a library of programming functions mainly
aimed at real-time computer vision. Originally developed by Intel, The Library is cross-platform and
free for use under open-source Apache 2 License. Application of OpenCV are Object detection,
Segmentation and Recognition and many more. OpenCV is written in C++ and its primary interface
is in C++ and there are bindings in Python, Java and MATLAB/OCTAVE. The API for these
interfaces can be found in the online Documentation.

Fig 3.1 License Plate detection using OpenCV

In this approach, OpenCV Image Augmentation functions were used to highlight the license plate
followed by contouring of License Plate bounding box. Figure 3.1 shows the major steps in object
detection using this method where License plate is detected based on obtained geometric coordinates
& extraction of the rectangular bounding box. This model is easy to implement but has limited
application for light color vehicles and lacks required accuracy
Ganesh Kumar P – 1CR19EC402 Dept. of ECE – Feb – June 2022 Page 5
Computer Vision based License Plate Detection for Automated Vehicle Parking Management System

3.2 HAAR Cascade

HAAR Cascade classifiers are an effective way for object detection. This method was proposed by
Paul Viola and Michael Jones in their paper rapid Object Detection using a Boosted cascade of
Simple Features. HAAR Cascade is a Machine Learning-based approach where a lot of positive
and negative images are used to train the classifier.

Positive Images – These images contain the images which we want our classifier to identify.
Negative Images – Images of everything else, which do not contain the object we want to detect.

HAAR cascade is an object detection algorithm with four stages viz – HAAR Feature selection,
creating integral images, Ada-boost training and Cascading Classifiers. It is used to identify objects
in an image or video and is based on the concept of features. It is well known for being able to detect
faces and body parts in an image, but can be trained to identify almost any object.

Fig 3.2 License Plate Detection using HAAR Cascade

The custom model was obtained by training the base model on Indian license plate. For character
recognition of the contents of License Plate, CNN model was used but due to low accuracy,
Pytesseract has been finally used. The result of prediction is shown in Fig 5. Though detection through
HAAR Cascade is faster but it has lower accuracy.

Ganesh Kumar P – 1CR19EC402 Dept. of ECE – Feb – June 2022 Page 6

Computer Vision based License Plate Detection for Automated Vehicle Parking Management System

3.3 YOLOv3
YOLOv3 (You Only Look Once, Version 3) is a real-time object detection algorithm that identifies
specific objects in videos, live feeds, or images. YOLO uses features learned by a deep convolutional
neural network to detect an object. Versions 1-3 of YOLO were created by Joseph Redmon and Ali
Farhadi. The first version of YOLO was created in 2016, and version 3, which is discussed
extensively in this article, was made two years later in 2018. YOLOv3 is an improved version of
YOLO and YOLOv2. YOLO is implemented using the Keras or OpenCV deep learning libraries.

Fig 3.3 Yolov3 Network Architecture

This is the selected model for the proposed system due to higher accuracy and model robustness. Here
the base model (YOLO- You Only Look Once) is built using transfer learning from the framework
DARKNET. As License plate is not part of coco dataset, model had been trained to detect this custom
object. YOLOv3 is another variant of YOLO. It is derived from YOLOv2 and DARKNET
Architecture.

The most salient feature of v3 is that it makes detections at three different scales. The detection is
done by applying 1 x 1 detection kernels on feature maps of three different sizes at three different
places in the network. First, YOLO v3 uses a variant of Darknet, which originally has 53-layer
network trained on ImageNet. For the task of detection, 53 more layers are stacked onto it, giving it
a 106 layer fully convolutional underlying architecture for YOLOv3.

YOLOv3 performs at par with other state of art detectors like RetinaNet, while being considerably
faster, at COCO mAP 50 benchmark. It is also better than SSD and its variants. However,
performance drops significantly as the IOU threshold increases. YOLOv3 has relatively high
performance (AP) for smaller objects and performs much faster than other models, it provides an
optimum solution for license plate detection.

Ganesh Kumar P – 1CR19EC402 Dept. of ECE – Feb – June 2022 Page 7

Computer Vision based License Plate Detection for Automated Vehicle Parking Management System

Fig 3.4 Comparative assessment- Yolov3 vs other models (mAP 50)

To develop the model for License Plate detection, we used transfer learning to train the model on
300+ custom car images with annotated License Plate and obtained the new weights for the updated
model. The model was tested for 20+ images and License Plate details are extracted. A blob was
constructed from the input image and passed to the model which gives the output image with the
bounding box identifying the license plate. The model was also tested on video clips which is needed
for final integration in Vehicle Parking Management System. At entrance video is read at 45 FPS for
identification of keyframe image and detected License Plate detail was transferred to Central Data
Management System for further processing.

Fig 3.5 License Plate detection using YOLOv3

Ganesh Kumar P – 1CR19EC402 Dept. of ECE – Feb – June 2022 Page 8

Computer Vision based License Plate Detection for Automated Vehicle Parking Management System

3.4 CHARACTER RECOGNITION

OpenCV had been used to read and augment the extracted license plate from all the above LP
detection models. By default, OpenCV stores images in BGR format and since Pytesseract assumes
RGB format, hence images were converted from BGR to RGB format. Different preprocessing image
augmentation techniques like Median blur, grayscale and bilateral filter were applied by using CV2
libraries.

Fig 3.6 Extracted License Plate by detection models.

Fig 3.7 Augmented Images

Python-tesseract is a python wrapper for Google's Tesseract-OCR used for character recognition from
images. Pytesseract is been used for text detection & recognition from augmented image output. This
method for character recognition has given better results than the CNN method for most of the images

Fig 3.8 License Plate Text Recognition

Ganesh Kumar P – 1CR19EC402 Dept. of ECE – Feb – June 2022 Page 9

Computer Vision based License Plate Detection for Automated Vehicle Parking Management System

3.5 PARKING SLOT MANAGEMENT

An effective parking space management necessitates the consolidation of used and free slots
information at various parking bays and their integration with CDMS for reporting and displays. In
the proposed system, the availability and usage of parking slots is reported by using the developed
YOLOv3 model. This model generates the information about free parking slots by identifying the
parked cars as object and then deducing the free slots using total available slots in that area. Figure
4.1 shows the schematic for parking management where the final information about free slots at each
level is displayed in LED boards which can be used to achieve an automated guided parking and
navigation.

Fig 3.9 Free Parking slot detection and display

Ganesh Kumar P – 1CR19EC402 Dept. of ECE – Feb – June 2022 Page 10

Computer Vision based License Plate Detection for Automated Vehicle Parking Management System

Chapter 4
RESULT & DISCUSSION
Figure 4.1 shows relative prediction accuracy and execution time of different models. The proposed
models were evaluated in two parts:

➢ Detection of vehicle's license plate - In order to evaluate the relative merits and assess the
performance of three different object detection models viz HAAR Cascade, OpenCV and
YOLOv3, models were trained on more than 300+ images and tested on 20+ car images covering
various data capture scenarios. YOLOv3 model has shown 100 % accuracy in LP detection and
Free parking slot detection which is significantly higher than other two models.

➢ Recognition of License Plate Number - For character recognition Pytesseract produced better
results with 90 % prediction accuracy as compared to CNN algorithm.

Fig 4.1 Accuracy and timing results for used models

Ganesh Kumar P – 1CR19EC402 Dept. of ECE – Feb – June 2022 Page 11

Computer Vision based License Plate Detection for Automated Vehicle Parking Management System

Chapter 5
CONCLUSION

In this paper, the development of Automated vehicle license plate detection system is highlighted
which has wide applications in many spaces viz Parking system in commercial spaces, traffic control
system, security measures through vehicle number tracking etc. To achieve the object
detection/recognition with high accuracy and precise results, various models like HAAR Cascade,
OpenCV and YOLOv3 had been studied and incorporated followed by text recognition using
Pytesseract. In order to develop the robust model with best results, YOLOv3 (with Open CV) had
been considered due to its superior performance This model when productionized by integrating with
other applications can be implemented for an automated guided parking and payment system in
commercial spaces like shopping malls, business parks etc., it will be a highly cost effective and easy
to implement solution requiring minimal infrastructure changes, saving time and efforts for Vehicle
Parking Management System (VPMS)

Ganesh Kumar P – 1CR19EC402 Dept. of ECE – Feb – June 2022 Page 12

Computer Vision based License Plate Detection for Automated Vehicle Parking Management System

REFERENCES
➢ [1] Narayana Darapaneni, Kiran Mogeraya, Shubham Mandal, Ashwin Narayanan, Prasanna Siva,
Anwesh Reddy Paduri, Faisal Khan, Praful Mohan Agadi, ” Computer Vision based License Plate
Detection for Automated Vehicle Parking Management System”, in 11th IEEE Annual Ubiquitous
Computing, Electronics & Mobile Communication Conference (UEMCON), 2022
➢ [2] SherrZheng Wang, HsMian Lee, “Detection and Recognition of License Plate Characters with
Different Appearences”, in proc. Conf. Intelligent Transportation Systems, vol. 2, pp. 979-984, 2003.
➢ [3] Preemon Rattanathammawat, Thanarat H. Chalidabhongse, “A Car Plate Detector using Edge
Information”, in proc. International Symposium on Communications and Information Technologies,
ISCIT '06., pp.1039-1043, October 18-September 20, 2006
➢ [4] X. He, H. Zhang, W. Jia, Q. Wu, T. Hintz, ‘Combining Global and Local Features for Detection of
License Plates in a Video’, in proc. of Image and Vision Computing New Zealand 2007, pp. 288–293,
Hamilton, New Zealand, December 2007
➢ [5] Otsu, N., A threshold selection method from gray-level histogram, in journal on IEEE Transactions on
Systems Man and Cybernetics, SMC-9, pp.62-66. 1979.
➢ [6] Qadri, M. T. and Asif, M. Automatic Number Plate Recognition System For Vehicle Identification
Using Optical Character Recognition; 2009 in proc. International Conference on Education Technology
and Computer. IEEE Xplore; pp.335-338. 2009.
➢ [7]. M. M. Rashid, A Musa, M. Ataur Rahman, N. Farahana and A. Farahana, “Automatic Parking
Management System and Parking Fee Collection Based on Number Plate Recognition” , International
journal of Machine Learning and Computing , Vol. 2, No. 2, April 2012, pp 93-9
➢ https://viso.ai/deep-learning/yolov3-overview/
➢ https://www.geeksforgeeks.org/python-haar-cascades-for-object-detection/
➢ https://towardsdatascience.com/face-detection-with-haar-cascade-727f68dafd08
➢ https://en.wikipedia.org/wiki/OpenCV
➢ https://github.com/opencv/opencv/wiki/Deep-Learning-in-OpenCV

Ganesh Kumar P – 1CR19EC402 Dept. of ECE – Feb – June 2022 Page 13