ThesisPDF Available

The enhancement of public building security service delivery through a biometric system that employs facial and pattern recognition techniques

Authors:

Abstract and Figures

In any building, security is a major concern; this is especially true in public buildings where commercial and government operations take place. To ensure security in public buildings, several access control mechanisms have been created over time around the world and these include a paper based logging system currently in use at most public buildings in Zimbabwe. This study aims to offer a solution for streamlining the check-in and checkout procedure for visitors at a public building site using a device with security and customer service goals that employs facial recognition and pattern recognition to identify people as they enter and log their entrance time. The primary focus is on solving real problems hence the researcher employed an action research approach. The deep learning facial recognition model was developed using Tensorflow and the pattern recognition model was implemented using a convolutional neural network. A simulation model to explore and demonstrate how the overall research proposal would work was developed using ReactJS, NodeJS and MongoDB. Only 47.06% of public buildings in Gweru have an access control mechanism at a security checkpoint. Of these 47.06 public buildings not a single building is making use of an electronic based visitors logging and auditing system yet crimes are prevalent at these buildings. The average time to process a person’s identity from their face at a security checkpoint is 9 seconds and that improves service delivery. Plastic ID cards are the identity cards that have over 80% accuracy when attempting to find a person’s identity from documents.
Content may be subject to copyright.
FACULTY OF ENGINEERING AND BUILT ENVIRONMENT
MSc Mechatronics and Artificial Intelligence
The enhancement of public building security service
delivery through a biometric system that employs
facial and pattern recognition techniques
BY
DZINAISHE MPINI (R211168H)
2022
SUPERVISOR: Dr P. MANYERE
i
Abstract
In any building, security is a major concern; this is especially true in public buildings where
commercial and government operations take place. To ensure security in public buildings,
several access control mechanisms have been created over time around the world and these
include a paper based logging system currently in use at most public buildings in Zimbabwe.
This study aims to offer a solution for streamlining the check-in and checkout procedure for
visitors at a public building site using a device with security and customer service goals that
employs facial recognition and pattern recognition to identify people as they enter and log
their entrance time. The primary focus is on solving real problems hence the researcher
employed an action research approach. The deep learning facial recognition model was
developed using Tensorflow and the pattern recognition model was implemented using a
convolutional neural network. A simulation model to explore and demonstrate how the
overall research proposal would work was developed using ReactJS, NodeJS and MongoDB.
Only 47.06% of public buildings in Gweru have an access control mechanism at a security
checkpoint. Of these 47.06 public buildings not a single building is making use of an
electronic based visitors logging and auditing system yet crimes are prevalent at these
buildings. The average time to process a person’s identity from their face at a security
checkpoint is 9 seconds and that improves service delivery. Plastic ID cards are the identity
cards that have over 80% accuracy when attempting to find a person’s identity from
documents.
ii
Acknowledgments
The list of people who made it possible for me to finish this research is long. It starts of course
with my supervisor, Dr P. Manyere who took his time to assist me with knowledge and
guidance on writing academic research. Our Mechatronics and Artificial Intelligence
programme coordinator and lecturer Dr T Mushiri’s encouragement, inspiration and support
has also been remarkable. He has always done his best to make sure we have all the desired
resources and support to finish the research. I also would like to extend my gratitude to
Benedict Chadiwa who motivated me to start the course and for providing guidance in
Mechatronics courses.
Among the people who inspire me in my education few people standout and these are:
My mother, my brother Ransome, my boss Dr T. Magadzire, my colleagues Kudakwashe and
Shingirayi and my friends Beverly, Allan and Leonard. I am really grateful for your support
and encouragement throughout the time I was doing this course.
iii
Dedication
To my loving mother Winfilder Mpini.
iv
Declaration
I Dzinaishe Mpini declare that this dissertation meets the rules and regulations that are in
the faculty guidelines. This dissertation is a result of my research and my investigations.
____________________________________ ____________________
Student Signature Date
v
Table of Contents
Abstract ...................................................................................................................................................................... i
Acknowledgments ................................................................................................................................................ ii
Dedication ............................................................................................................................................................... iii
Declaration ............................................................................................................................................................. iv
Table of Contents .................................................................................................................................................. v
List of Figures ..................................................................................................................................................... viii
List of Tables .......................................................................................................................................................... ix
Abbreviations ..........................................................................................................................................................x
Chapter 1 - Introduction ..................................................................................................................................... 1
1.1 Introduction ........................................................................................................................................... 1
1.2 Background ........................................................................................................................................... 1
1.3 Problem Statement ............................................................................................................................. 2
1.4 Research Aim ........................................................................................................................................ 2
1.5 Research Objectives ............................................................................................................................ 2
1.6 Research Questions ............................................................................................................................. 3
1.7 Research Hypothesis ........................................................................................................................... 3
1.8 Justification ............................................................................................................................................ 3
1.9 Scope of Study ....................................................................................................................................... 4
1.10 Limitations .......................................................................................................................................... 4
1.11 Definition of Terms ........................................................................................................................... 5
1.12 Dissertation Flow .............................................................................................................................. 6
Chapter 2 - Literature Review .......................................................................................................................... 7
2.1 Introduction ........................................................................................................................................... 7
2.2 Security and public buildings .......................................................................................................... 7
2.3 Security Control Methods ................................................................................................................. 8
2.4 Physical Access Control Mechanisms ........................................................................................... 9
2.5 Logging ...................................................................................................................................................10
2.5.1 Paper Based Visitor Log System ......................................................................................10
2.5.2 Electronic Visitor Management Systems .....................................................................11
vi
2.5.3 Smart Card and RFID Based Visitor Management Systems ..................................12
2.5.4 Biometric Based Logging Systems ..................................................................................12
2.6 Research Gaps .....................................................................................................................................17
Chapter 3 - Materials and Methods ..............................................................................................................19
3.1 Introduction .........................................................................................................................................19
3.2 Nature of Research ............................................................................................................................19
3.3 Ethical Considerations .....................................................................................................................21
3.4 Research Design ..................................................................................................................................22
3.4.1 Quantitative Approach ........................................................................................................22
3.4.2 Qualitative Approach ...........................................................................................................22
3.5 Model Development ..........................................................................................................................30
3.6 Conclusion .............................................................................................................................................33
Chapter 4 - Results and Discussion ..............................................................................................................34
4.1 Introduction .........................................................................................................................................34
4.2 Access Control Mechanisms being used in Gweru buildings ............................................34
4.3 Crime records most prevalent on public buildings between April 2021 and March
2022 ................................................................................................................................................................35
4.4 Facial Recognition Model Accuracy and Time Taken Results ...........................................36
4.5 Pattern Recognition Model Accuracy Results .........................................................................36
4.6 Conclusion .............................................................................................................................................37
Chapter 5 - Recommendations and Conclusion ......................................................................................38
5.1 Introduction .........................................................................................................................................38
5.2 Aims and Objectives Realisation ..................................................................................................38
5.3 Challenges Faced ................................................................................................................................38
5.4 Recommendations for future work .............................................................................................39
References .............................................................................................................................................................41
Appendices ...................................................................................................................................................49
Appendix A: Manual log sheet .....................................................................................................49
Appendix B: Access control mechanisms in different public buildings in Gweru
Observations Score Sheet .............................................................................................................50
Appendix C: Interview Guideline for Public Buildings .....................................................51
vii
viii
List of Figures
Figure 2.1: Architecture of a biometric system .............................................................................13
Figure 3.1: Action Research Phases ....................................................................................................20
Figure 3.2: Face variants of one person object under study .....................................................23
Figure 3.3: Face recognition processing flow. ................................................................................24
Figure 3.4: Facial recognition made on a clear face .....................................................................25
Figure 3.6: Facial recognition made on a face with a face mask on chin .............................26
Figure 3.7: Zimbabwean Passport ......................................................................................................27
Figure 3.8: Zimbabwean Paper Drivers Licence ............................................................................28
Figure 3.9: Zimbabwean Metal Drivers Licence ............................................................................29
Figure 3.10: Zimbabwean Plastic ID Card ........................................................................................30
Figure 3.11: Security Checkpoint Simulation System Facial Recognition Screen ............32
Figure 3.12: Security Checkpoint Simulation System Initiate Pattern Recognition Screen
.................................................................................................................................................................32
Figure 3.13: Security Checkpoint Simulation System Pattern Recognition Screen .........32
Figure 3.14: Security Checkpoint Simulation System Event Success Screen .....................33
ix
List of Tables
Table 4.1: Access control mechanisms in use in public buildings ..........................................34
Table 4.2: Crime records performed on building between April 2021 and March
2022 ......................................................................................................................................................35
Table 4.3: Results from human objects studied .............................................................................36
Table 4.4: Results on accuracy of pattern recognition model on identifying identity
documents patterns ........................................................................................................................37
Abbreviations
CBD - Central Business District
CCTV - Closed Circuit Television.
CNN - Convolutional Neural Network
DL - Deep Learning
ML - Machine Learning
MTCNN - Multi Task Cascaded Convolutional Neural Network
RFID - Radio Frequency Identification
SSSM - Second Street Shopping Mall
1
Chapter 1 - Introduction
1.1 Introduction
In any building, security is a major concern; this is especially true in public buildings where
commercial and government operations take place. To ensure security in public buildings,
several access control mechanisms have been created over time around the world. This
study aims to offer a solution for streamlining the check-in and checkout procedure for
visitors at a public building site using a robot with security and customer service goals that
employs facial recognition and pattern recognition to identify people as they enter and log
their entrance time.
The researcher will offer a background for the research in this chapter, detailing the breadth
of the problem that the research aims to solve. This chapter will also define the research's
goal, objectives, and scope. The researcher will define the common terms found in this
research before the end of this chapter and finally discuss the dissertation flow.
1.2 Background
Security is a major concern in every building; this is especially true for public buildings
where business and government activities take place. Different access control mechanisms
have been developed over the years to enforce security in public buildings (Kosar and
Ahmed, 2000; Karimah et.al, 2007; Chung and Taehwang, 2018; Okamura & Norimura, 2014;
Wan et.al, 2021). These include;
the keeping and updating of a log book in the custody of security personnel which is
filled by visitors as they enter and exit a building (Oktaviandri and Keat, 2018; Ghaithi
and Eaganathan, 2016),
electronic based visitor management systems which are in short electronic visitor log
books (Chaudhary et.al, 2021)
fingerprint systems (Isa et.al, 2010)
facial recognition systems (Satari et.al, 2014; Gautam et.al, 2021)
The facial recognition technique is the most efficient human recognition technique as
outlined by Sharma (2017).
2
In Zimbabwe at most public buildings visitors activity records are managed through a log
book. An example of records kept in such a log book is shown in Appendix A. The most
common records are first name, last name, national registration number, time in, time out
and signature. The researcher carried out a survey of how different public buildings in
Zimbabwe are managing security and the results showed 43% maintained a visitor’s
logbook, 53% had no logbooks. The findings of the survey are summarised in Appendix B.
This manual visitor’s log system has notable concerns of privacy, service delivery and
security enforcement. Motivated by the inefficiency of the manual visitors log system still in
use in most public buildings in Zimbabwe, this research seeks to propose a system to
streamline the checkin and checkout process for visitors at a public building site through a
robot with both security and customer service goals that uses computer vision to identify a
visitor as they come in and log their entrance time. If the robot fails to identify the person it
will then ask the person to scan an identity document so as to capture their details or to
capture a new variation of the visitor's face.
1.3 Problem Statement
Current access control systems employed by most buildings in Zimbabwe have notable
concerns of visitors’ privacy as anyone can see what has been logged in a book as they will
be logging. Again service delivery using the current method is slow and there is poor security
enforcement as a visitor can easily fake identification.
1.4 Research Aim
To develop a security checkpoint assistant robot armed with computer vision and pattern
recognition techniques so as to quickly serve customers in a safe way whilst at the same
enforcing more efficient building security.
1.5 Research Objectives
1. To streamline a public building visitors checkin and checkout process through facial
recognition.
2. To register first time customers through pattern recognition done on national
identity documents, passports or driver’s licence.
3. To screen visitors entering a building against a company’s security watch list thereby
allowing selective access to a building.
4. To provide insights and analytics on building visitors activity.
5. To pre register guests in the case of many visitors who will visit the building at one
time.
3
1.6 Research Questions
1. What access control mechanisms are currently being used in Gweru?
2. Which crimes are most prevalent on public buildings in Gweru?
3. How best can computer vision (facial recognition) be merged with pattern /
character recognition techniques so as to improve security checkpoints service
delivery whilst at the same time maintaining the security goal of security
checkpoints?
1.7 Research Hypothesis
H1 - Facial recognition merged with character recognition techniques are efficient in
improving service delivery whilst maintaining the security goal of security checkpoints.
H0 - The current manual methods on managing security points have better service delivery
and have better security goals in comparison to the proposed method.
1.8 Justification
This idea presents a technological solution for improving public building security while also
ensuring speedy and effective service delivery and reducing visitor annoyance. Visitors,
renters, and security employees at Second Street Shopping Mall will benefit greatly from the
proposal's successful execution and results. It's possible that the project will be expanded to
include other commercial real estate players.
To security personnel the device will help to quickly do check in and checkout processes on
visitors. The proposed system reduces the operations burdens on the front desk as it
automates most operations and procedures. This in turn will help reduce the number of
guards required for manning a building as well as reducing the labour force cost. The
technology will also help security personnel in dealing with piggy-backing visitors, as the
device, unlike humans, will treat all visitors the same and not be selective to bribes and
emotions.
To visitors, the technology will reduce visitor waiting times for logging in their details every
time they cross through a security checkpoint. The device will do the identification and
logging automatically. The proposal also ensures the privacy of every visitor is maintained
compared to a manual logbook where everyone who comes after you can view who else is in
the building and again potentially grab people's contact details illegally.
To tenants and real estate players the technology will be of benefit in that it enhances
building security and protects the workplace from potential security threats as it screens
against a watch list of criminals and miscreants as the system raises an alarm when one has
4
been identified. Thus their property will always be better protected. Unlike humans the
technology will be efficient in logging every entrance and exit hence when an incident such
as theft occurs; it will be easier to check who was in the building at the time. Moreover the
proposal leverages technology for a safer workplace especially in the wake of some
pandemics such as Covid 19 where avoiding contact is encouraged.
1.9 Scope of Study
This research was conducted on Gweru public buildings and will only cover facial
recognition and pattern recognition on identity documents.
1.10 Limitations
Nowadays, because of Covid 19, people are required to move around with face masks so as
to reduce the chances of spreading the virus. The proposal relies on facial recognition which
works best when someone does not have a face mask. Researchers around the world
however have started proposing and developing algorithms that can detect a face with a face
mask on (Song et.al, 2021). Since this is a new technology it may be poor in achieving its task
but with continual improvement facial recognition in a masked face can be achieved.
People from developing countries are also generally reluctant in adapting to technology
innovations and advancement (United Nations Report on Trade and Development, 2021).
This may pose a challenge in motivating the development of the technology when it will not
be accepted. The solution to this would be to work with all stakeholders of the proposal from
the start so that they are involved in the development process and that may ease the
adoption process. It is also best to identify the few willing tenants of some buildings to work
with until implementation. The other building owners may in turn realise the benefits of such
a system later and adopt it. Second Street Shopping Mall has already agreed to be used during
the test stages of this project.
The platform does not address the “way finding problem” which is a scenario where
someone enters a building but is unsure of the directions they should take to get to the office
or spot they need service from. Generally with a human, one can then explain where they
want to go to the front desk personnel and the front desk personnel explain how they can
get there. Upon successful implementation of the project, it would be best in the future to
add voice recognition and direction finding techniques to the device such that a visitor can
then explain where they want to go and the device can direct them.
Lastly the proposed technology cannot classify visitors and their purposes. Some classes that
already exist that may be put in place include family, friends, tenants, business visitors and
service providers such as delivery guys and cleaning services personnel. On classification, it
is recommended to then have a feature that classifies each visitor for analytics purposes.
5
1.11 Definition of Terms
Access control - A security checkpoint system consists of a human stationed at a fixed
position within a premise to control the entrance and exit of people, goods and items.
Biometric system - a system that uses mathematical algorithms and biometric data to
recognise a certain trait of an individual.
CCTV - closed circuit television.
Facial recognition - a type of biometric software which creates a mathematical map of a
person's facial features and stores the information as a faceprint.
MongoDB - open-source, cross-platform document-oriented database programme.
MTCNN - Multi Task Cascaded Convolutional Neural Network. (Face detection neural
network used to detect faces and facial landmarks on pictures).
Node.js - network application framework that employs an asynchronous event-driven
JavaScript engine to develop scalable network applications
OpenCV - Python library that provides a real time computer vision interpretation.
Pattern recognition - refers to the use of machine learning techniques and algorithms to
identify patterns supplied.
Public building - any building where government and or commercial activities take place which
allows access to different visitors in the building.
Python - high level programming language with very simple syntax.
React - a JavaScript-based user interface library
6
Security - the control of access into premises, organisations, buildings, rooms and spaces.
RFID - Radio Frequency Identification
Tensorflow - framework for machine learning specifically designed for neural networks.
Wayfinding problem - a scenario where someone enters a building but is unsure of the
directions they should take to get to the office or spot they need service from.
1.12 Dissertation Flow
The remaining sections of this paper are organised as follows:
Chapter 2 - Literature Review: This chapter examines, analyses, and synthesises pertinent
research documentation on authentication, authorization, security practices and
requirements, access control models and implementations, and other supporting
technologies that will be beneficial in achieving the study objectives.
Chapter 3 - Materials and Methods: This chapter provides a design for an attribute-based
access control system as a proof of concept. The modular library structure is introduced,
along with various success criteria, design concerns, and alternative techniques.
Chapter 4 - Results and Discussion: This chapter offers the system's output as well as an
analysis of it in relation to the success criteria stated in Chapter 3.
Chapter 5 - Recommendations and Conclusion: This chapter uses the analysis, discussion,
and results from Chapter 5 to draw conclusions about the research and compare them to the
research objectives set forth in this chapter. This chapter also discusses prospective
extensions or solutions to the research's shortcomings.
7
Chapter 2 - Literature Review
2.1 Introduction
This chapter, the literature review chapter, is concerned with providing background to the
public buildings security situation and providing evidence for the need of an intelligent agent
to improve service delivery at a security checkpoint. The researcher will define security,
outline its genesis and the importance of security in public buildings. Different types of
security methods used in public buildings will be described and finally the researcher will
narrow down the access control mechanisms in public buildings singling out the ones being
used in Zimbabwe. In conclusion the researcher will debate the need for better access control
mechanisms at a public building.
2.2 Security and public buildings
Phin et.al (2020) defines security as the control of access into premises, organisations,
buildings, rooms and spaces. It is a very old concept which seeks to protect the assets and
belongings of an individual, organisation, company or group of people. Security is a critical
concern around the world (Sinha et.al, 2015). When applied in public buildings it refers to
the control of access into organisations, buildings, rooms, and information technology (IT)
peripherals (Phin et.al, 2020). In this research a public building is any building where
government and or commercial activities take place which allows access to different visitors
in the building.
The goals of security are to protect and preserve assets' value. Phin et.al (2020) and
Challinger (2008) outlines the problems with lack of physical security and these include
vandalism,
theft and burglary,
terrorism,
intrusion,
assets loss,
assets and equipment damage
system disruptions,
8
data breaches,
offences against persons which include general violence and ugly confrontations
unauthorised access to utilities such as electricity, telephones and internet
disorder and drug affected behaviour.
According to Mawby (2014) the major security problem with public buildings is theft and
burglary. Mawby also reported that of the theft and burglary crimes that occur on corporate
property worldwide, most of them are conducted on commercial premises, hence there is a
valid concern on the need to improve commercial buildings security. Mawby further
reported that intruders and breaking into business premises was considered an area
problem of 36% of businesses. 67% businesses think it has a detrimental effect on their
businesses.
In Zimbabwe Safeguard (2021) reported that intruder invasions account for 15.38% of
robbery, hijacking, break in, invasion and theft crimes. The report however does not further
breakdown the incidents classifying if they were on residential, commercial or government
buildings.
2.3 Security Control Methods
Security control methods are techniques employed to enforce security measures inside a
defined structure to dissuade or prohibit unauthorised access to sensitive materials, places,
structures, and people (MIT, 2022). Access control, surveillance, and testing are the three
basic components of the physical security framework. How well each of these components is
implemented, enhanced, and maintained typically determines the success of an
organisation's physical security programme.
Government of District Columbia (2012) outlines the following security control methods
that can be implemented in public buildings;
1. Planning and preparedness which includes threat analysis, vulnerability assessment,
consequences analysis, outlining procedures, and security audits,
9
2. Conducting background checks on employees and tenants to limit number of people
who can potentially breach security expectations whilst inside a building
3. Access control mechanisms which emphasise on only allowing access to vetted
people
4. Establishing perimeter barriers around the building facility
5. Developing a process for communicating to building management, employees,
tenants about the current security situation as well as security awareness training.
6. Monitoring, surveillance and inspection
7. Developing and implementing a security plan for computer and information systems
8. Establishing an incident response plan
This research focused on the third technique which mentions an access control mechanism
technique and more specifically the security checkpoints in buildings, hence it is the only
method that will be explained in detail in the coming section.
2.4 Physical Access Control Mechanisms
In the past, and in some cases still today, businesses depended solely on perimeter security,
which is the process of routing all human traffic through a single or very few entrance and
exit points where they are scanned and security policies are implemented (Linklater, 2016).
This gave birth to the idea of security checkpoints. A security checkpoint system consists of
a human stationed at a fixed position within a premise to control the entrance and exit of
people, goods and items. Sandhu and Samarati (1994), Archibald et.al, 2002, Forte (2006,
2009) and Linklater (2016) commented that establishing a perimeter security alone is not
enough itself and therefore suggests additional systems to improve security.
Over time the systems for perimeter security and security checkpoints have evolved and are
being improved with each iteration. These system iterations include addition of paper based
visitors log books, electronic visitors log systems, closed circuit television (CCTV)
monitoring, facial recognition techniques and the use of smart or RFID cards. The next
section makes a critical analysis of different access control mechanisms currently being used
in the world.
10
2.5 Logging
Sandhu and Samarati (1994) proposed logging as an additional system to improve perimeter
security. In the context of security, logging refers to the recording of events at a security
checkpoint. The goal of logging, according to Sandhu & Samarati (1994), is to pair it with
auditing, which he defines as "a posteriori analysis of all activities." As a result, the activities
must be recorded for later study. Forte (2009) defines auditing as the process of discovering
abnormalities, while Casey (2008) characterises a complete security solution for a system as
follows:
Establishing Goals: Establish a resource security policy.
Enforcing Targets: Use access control to restrict access to the resource based on the
security policy and the subject's rights.
Verifying Compliance: Through the auditing process, ensure that the security policy
is being followed correctly.
In order to achieve these goals Casey (2008) and Forte (2006) specify that logs must contain
the following or follow the following standards:
Security logs must conform to standards necessary to be provided in law courts
Security logs must be drawn in a way that they show a particular incident under
question
Security logs must show the complete incident or activity
Security logs must not be tampered with and altered
Security logs must have time stamps
2.5.1 Paper Based Visitor Log System
The desire to achieve these goals led to the adoption of visitors' log sheets or books referred
to as paper based visitor log systems in this research. A paper based visitor log is a book in
security personnel's possession at a security checkpoint that is filled out by visitors when
they enter and exit a building (Ghaithi and Eaganathan 2016; Oktaviandri and Keat 2018).
As observed by the researcher most logbooks contain the following pieces of information on
each single entry.
11
Date of visit,
Visitor’s full name,
Visitor’s national registration number,
Visitor’s address,
Visitor’s phone number,
Visitor’s purpose of visit,
Entrance time and
Exit time.
A single entry in the book signifies entrance and exit made by a visitor at a building.
As alluded earlier logging visitors has advantages in that it allows screening and controlling
selective access into a building and in cases of an incident such as theft one can go to the
records and find out who was in the building at the time and where they were based on the
information supplied as they logged entrance, but the entries are manual entries and are
made on paper, therefore this method has a number of disadvantages especially in modern
society. Ghaithi and Eaganathan (2016), Oktaviandri and Keat (2018) and Chadhary et.al
(2021) outline the following disadvantages with manual and paper based visitor logs:
Exhibit privacy concerns as a visitor, whilst entering their own details can see details
already entered by other visitors
Consumes longer time in entering records
Lack of methods to verify and authenticate each visitor
Inadequate tracing as it is difficult to analyse paper based records
2.5.2 Electronic Visitor Management Systems
In light of the limitations of the manual and paper based visitor logs, some buildings have
started using electronic based visitor management systems. An Automated Visitors
Management System was introduced by Bhandari et al. (2010). The system tracked how a
public building or site was used, obtaining information on each visitor and their location. The
technology worked in tandem with building security and access control systems. At security
checkpoints, the method allowed for rapid service delivery.
Gaithi and Eaganathan (2016) deployed a web based system that assists with logging in and
out of visitors, uploading personal details manually, reporting visitor activity in building and
sending SMS alerts and notifications if a security breach is suspected. The system was
considered much better than manually updating a log book.
12
Chaudhary et al. (2021) created a system that assisted visitor management by tracking
visitors, staff, assets, and supplies as they entered and exited the facility. The technology
eliminated visitor lines by processing numerous people at once at one station, allowing
tourists and employees to spend more time in the facility. The system kept track of visitor
passes, expiration dates, visit locations, hosts visited, and the purpose of the visit. All of this
data was kept in a database. According to Chaudhary et al. (2021), the system generated end-
of-day reports to assure regulatory compliance. The system, however, depended on visitor
passes and did not check that the real person entering with a pass was the same as the person
on the pass.
Rivera (2021) presented a mobile-based visitor management system that would work in
conjunction with a computerised system. A visitor to a building uses the mobile application
to get visit notifications, respond to the computerised visitor management system, and
update their current location within the facility. It can also assist a user in locating their
desired office within a building.
The electronic visitors management systems are advantageous in comparison to the paper
based visitor log systems in that they serve customers quicker and have better analytics
capabilities but they still require human involvement to man a security checkpoint. In
addition, the systems do not prove the authenticity of the visitors entering a security
checkpoint.
2.5.3 Smart Card and RFID Based Visitor Management Systems
In 2018, Oktaviandri and Keat improved the concept of electronic based visitor management
systems through development of a system which extracts information from the Government
of Malaysia multipurpose card. The system resulted in quicker serving of building visitors
and reduced workload.
Gautam et.al (2021) also worked on a visitor management system which makes use of a
smart card and with better visitor tracking features. These two developments however
lacked the capabilities to authenticate a visitor’s identity. Information was obtained by
simply copying data on the card and entering in the developed systems.
2.5.4 Biometric Based Logging Systems
Researchers have also proved the efficiency of biometric systems in maintaining public
buildings security. Arulogun (2008) initiated the research which focused on keyless door
opening which relied on biometric recognition. A biometric system is a system that uses
mathematical algorithms and biometric data to recognise a certain trait of an individual
(Guennouni, Mansouri and Ahaitouf, 2020). Biometric systems can be used for a variety of
purposes. There are several systems that require users to enrol upstream. Fingerprint, face
and voice recognitions are all part of biometric systems helping in the identification and
authentication of humans.
13
A biometric system works in five stages as shown on Figure 2.1 below:
Figure 2.1: Architecture of a biometric system (Guennouni, Mansouri and Ahaitouf, 2020)
The capture module is the biometric system's entrance point, and it comprises capturing
biometric data in order to extract a digital representation.
The signal processing module allows for the optimization of processing time and digital
representation acquired during the enrollment phase in order to speed up the verification
and identification phases.
The biometric templates of system enrollees are stored in the storage module.
The matching module compares the data retrieved by the extraction module with the data
of the registered models to assess how similar the two biometric data are.
14
The decision or judgement module determines if the similarity index returned by the
matching module is sufficient for determining an individual's identity.
Biometric systems can be used for a variety of purposes. Biometrics can assist with
transactions for security concerns, making ordinary life safer and more practical. These
applications use cases include;
Justice and law enforcement
Border and airport control
Healthcare
Security
Finance
Mobile applications
Screen navigation
Aviation
In this research, the student mainly focuses on the security application of biometric systems.
The following sections seek to explore past work and research of biometric systems on
security with more interest in public buildings use cases.
2.5.4.1 Fingerprint Based Systems
Fingerprints are impressions made by the papillary ridges on the ends of the fingers and
thumbs. Fingerprints provide an impenetrable form of personal identification since the ridge
arrangement on each human finger is unique and does not alter with age or growth
(Alodainy, 2022). Biometric systems that are fingerprint based depend on these fingerprints.
Alphonse Bertillon, a Parisian police clerk, initiated work on fingerprint recognition after he
invented the Bertillon system in the late 1800s, which relied primarily on an elaborate set of
eleven meticulously recorded anatomical measurements - the length and width of the head,
the length and width of the left middle finger, the distance from the left elbow to the tip of
the left middle finger, and so on. Sir Francis Galton (the Galton point is named after him)
discovered the unique nature of human fingerprints when the Bertillon system was gaining
favour. In 1892, he created the masterpiece "Finger Prints." The UK's Home Ministry Office
15
declared in 1893 that no two people had the same fingerprints. (Usmani, Irum, Afzal and
Mahmud, 2013).
Clancy (2003) developed a system to improve building security by encrypting and storing
the private keys from fingerprint information on a smart card. One would then use the
smartcard to access restricted areas. Storing the key in a smart card was to solve the problem
of EMV loss before it’s decoded when entering the fingerprint directly on a fingerprint
capturing device.
Arakala (2009) provided the needs and techniques for developing a fingerprint biometric
authentication system that protects the fingerprint template throughout storage and
comparison. The thesis's principles created a solid framework for using fingerprint
information to authenticate individuals at security checkpoints in public buildings.
A fingerprint-based authentication system was proposed by Gao (2010). The system's user
must first register his or her fingerprints with a server database of fingerprint-minutiae
templates, which are created from the user's original fingerprint images and then populated
with a variety of foreign minutiae. The same live fingerprint will be measured again upon
entering a building to regenerate a template, which will then be checked against the saved
templates. Once the match is complete, the regenerated template will be discarded. Because
vast numbers of foreign elements are inserted into database templates, the system was
secure and private. Isa et.al (2010) also worked on a fingerprint system for the security
checkpoint of an entrance system into a building.This further strengthened the feasibility of
fingerprint based systems in security checkpoints systems.
The two most crucial things to consider while building a biometric system, according to
Gupta and Bhalla (2021), are security and recognition accuracy. As a result, they conducted
a thorough analysis of the most recent breakthroughs in the study of unique fingerprint-
based biometrics from these two perspectives, with the goal of improving system security
and identification accuracy. They also suggested combining fingerprints from different
layouts using a normalised fingerprint model. User Interface (UI) assaults and attacks on
16
format databases, according to the researchers, are two types of attacks on biometric
systems.
Given that, we can conclude fingerprint based systems are an efficient form of security
enforcement and still commonly used in private companies. However, fingerprint systems
are not recommended in these Covid-19 times where minimum contact with equipment and
other individuals is encouraged.
2.5.4.2 Facial Recognition Based Systems
Facial recognition has evolved as a solution to various current needs for identifying and
validating a person's identification. It satisfies the requirements of a biometric system, which
aims to recognise an individual's status by using traits unique to the body and functionalities
more familiar with the operation of visual surveillance (Mohsienuddin and Sabri, 2020).
The face recognition system works by computationally assessing the contours of the faces in
terms of their location and distance between a set of geometrical coordinates. The centre of
both pupils, the bridge of the nose, and the ends of the brows are all included in the
coordinates. Because each person is given a unique face print, the system makes a match
with the specified individual when the properties of sets of the geometry of an image that is
collected are compared to a database of a pre-existing recognisable image of a person. The
ability to verify a person's identification, as well as the technological forms that correlate to
recognising the face, have been developed to analyse and scan facial expressions in order to
determine a person's moods and feelings (Mohsienuddin and Sabri, 2020).
Past developments have been made on facial recognition to improve either security or
service delivery of buildings. Owayjan et.al (2013) initiated the work through development
of a facial recognition system with the main focus of detecting intruders and minimising
human error in vetting people to be allowed into a building. The system’s core functionality
however was to identify intruders entering a building.
17
Satari et.al (2014) furthered the use of facial recognition to improve security through the
2014 proposal to use facial recognition as an authentication technique in managing and
monitoring faces into a building. Farayola and Dureja (2020) further improved the facial
recognition concept through Deep Learning (DL) using Convolutional Neural Network (CNN)
architecture to ascertain the feasibility of DL in facial recognition. The model had a 97%
accuracy rate and can be applied to any camera footage system making it the best to apply
in a security setup. Keven (2021) worked on a system that improves service delivery at
security checkpoints and reduces chances of Covid-19 transmission. The system however
does not put emphasis on security as visitors say out their names and there is no way for the
device to verify the identity and or authenticity of a visitor. Gautam et.al (2021) proposed a
contactless visitor management system in the wake of Covid 19. Song et.al (2021) worked
on face in face masks which worked through detecting a face has a face mask worn,
classifying the mask, identifying the mask position and finally identifying the face wearing a
mask. All these proposals and developments however lack pattern recognition techniques
on identity documents so as to be able to identify a person.
In improving these facial recognition systems, autonomous and intelligent agents to serve as
security guards are also being developed by robotics engineering organisations and
researchers. Trovato et.al (2021) developed a humanoid to be employed in a real world
situation providing information about the area that it safeguards. Salamat et.al (2016)
compared robotic security guards with human security guards and concluded that robotic
security were more efficient than human security guards. Robot engineering firms have also
advanced the development of robotic guards development (Cobalt Robotics, 2021; SMP
Robotics 2021; Knightscope, 2021).
2.6 Research Gaps
No pattern recognition techniques applied to public buildings security research was found
but Artificial Intelligence researchers have proved the efficiency of pattern recognition
techniques on identifying characters on identity documents. (Naiguo et.al, 2015; Rayna and
Hanafiah, 2015; Castelblanco, 2020; Onespan2021).
The researcher could not find any work carried out in Zimbabwe on improving public
buildings security through electronic or biometric means. A survey carried out on buildings
in Gweru Central Business District (CBD) indicated none of the buildings had either an
electronic based or a security checkpoint system that employed facial recognition or pattern
recognition techniques. The summary of the survey is outlined in Appendix B. The researcher
therefore, made a survey on the public buildings security situation and needs in Zimbabwe
through onsite observations and a survey carried out on commercial buildings in Gweru
CBD.
18
19
Chapter 3 - Materials and Methods
3.1 Introduction
In this chapter, titled “Materials and Methods”, the researcher outlined how the research was
designed, justifying research decisions. This was made possible through outlining the nature
of the research, ethical considerations and the data collection processes that were used to
gather data for the study. Following that, a research design that clearly displays the
investigation's basic plan, strategy, and rationale was developed. This allowed the research
to be confirmed and to derive new conclusions from it. Finally, the researcher discussed the
deep learning technique and strategy that were used to construct the facial recognition and
pattern recognition model.
3.2 Nature of Research
The researcher proposed an action research approach. Action research is a method of
problem diagnosis and solution development in which the action researcher and the client
work together to diagnose the problem and build a solution based on the diagnosis (Bryman
and Bell, 2011). Simply said, action research is "learning by doing" - a group of people
identify an issue, fix it, assess their success, and try again if they aren't satisfied. While this is
the foundation of the technique, there are other crucial characteristics of action research that
set it apart from other types of problem-solving activities that we all do on a daily basis.
20
Figure 3.1: Action Research Phases
Because its primary focus is on solving real problems, action research is employed in real
scenarios rather than contrived, experimental studies. Social scientists, on the other hand,
can utilise it for preliminary or pilot research when the context is too vague to formulate a
specific study topic. Most of the time, however, it is chosen in accordance with its principles
when situations necessitate flexibility, participation of people in research, or rapid or holistic
change.
In general, there are three types of action research: positivist, interpretative, and critical.
Positivist action research, often known as "classical action research," considers research
to be a social experiment. As a result, action research has gained acceptance as a method for
testing ideas in a real world setting.
Interpretive action research also known as "modern action research," sees business reality
as socially created and focuses on local and organisational variables.
21
Critical action research is a sort of action research that takes a critical look at corporate
processes and looks for ways to enhance them.
This research shall make use of a critical action research approach. The reasons for applying
action in this research are based on the nature of action based research which is;
The business research has a high level of practical applicability;
It can be utilised with both quantitative and qualitative data;
Possibility of gaining a thorough understanding of the problem;
3.3 Ethical Considerations
When conducting research and gathering data, the researcher must consider whether the
procedures could injure the participants physically or emotionally. The researcher's ethical
behaviour was governed by the following principles:
To ensure that all necessary people, committees, and authorities have been contacted,
and that the work's guiding principles have been agreed upon in advance.
All participants must be given the opportunity to influence the work, and those who
do not choose to engage must have their wishes respected.
The progress of the project must be apparent and available to input from other
contributors.
Gathered sensitive and confidential information of people and their identity
documents will not be made public
To not watch or observe people’s behaviour at public buildings without their consent
To not pass on other people's knowledge as if it were their own. In other words, the
researcher must avoid plagiarising content.
To not expose sensitive information obtained from public buildings. This information
may be in the form of people’s sensitive data and company/building operating
procedures.
Before being published, descriptions of others' work and points of view must be
agreed with individuals involved.
22
3.4 Research Design
The purpose of the research design is to offer a suitable framework for a study (Creswell,
2013). The decision to be made about the research approach is a very important decision in
the research design process since it affects how relevant information for a study will be
gathered; yet, the research design process contains several interrelated decisions. This study
employed a mixed approach. The mixed methods approach combines numerical
measurement with in-depth investigation.
3.4.1 Quantitative Approach
Quantitative research methods focus on objective measurements and statistical,
mathematical, or numerical analysis of data acquired through polls, questionnaires, and
surveys, as well as modifying pre-existing statistical data using computing tools (Creswell,
2013). Quantitative research is concerned with collecting numerical data and generalising it
across groups of people or explaining a phenomenon. Quantitative research involves
gathering and analysing numerical data for statistical analysis.
This approach is beneficial in answering the research question “What access control
mechanisms are currently being used in Zimbabwe?”. Surveys and questionnaires will be
employed to investigate security access control mechanisms currently being used in
Zimbabwe. Of these, a survey was carried out to investigate security access control
mechanisms currently being used on Zimbabwean buildings. A questionnaire was carried
out to investigate crime records performed on buildings between April 2021 and March
2022.
3.4.2 Qualitative Approach
Qualitative research entails gathering and evaluating non-numerical data (e.g., text, video, or
audio) to comprehend concepts, opinions, experiences and gain in depth understanding of a
problem (Creswell, 2013). This approach is beneficial to address the research question “How
best can computer vision (facial recognition) be merged with pattern / character recognition
techniques so as to improve security checkpoints service delivery whilst at the same time
maintaining the security goal of security checkpoints?”.
23
In answering the research question, the researcher employed an explanatory research type,
where the researcher’s ideas and thoughts on facial recognition and pattern recognition
were explored on the topic of improving privacy and service delivery on security
checkpoints. The following four (4) phases were religiously followed to efficiently make the
research.
1. Data gathering of faces
2. Facial recognition model training and validation
3. Data gathering of identity documents
4. Pattern recognition training on identity documents
3.4.2.1Data gathering of faces
The data gathering of faces was important so as to train the facial recognition model on how
to identify unique features on someone’s face. The researcher observed after a thorough
search on the internet that there is no publicly available dataset of Zimbabwean faces. There
however exist a number of facial recognition databases publicly available for different
scenarios. The platform with the most datasets is https://www.face-rec.org/databases (Face
Recognition, 2021).
Of all the available datasets, the researcher noticed that these faces were not best for use in
Zimbabwe as they mostly do not apply to the Zimbabwe situation in terms of eye colour, hair
colour and other facial patterns common with most Zimbabweans. For that reason, a data
gathering of faces was conducted with 10 human subjects. On each face the researcher had
at least 10 poses and 30 illumination variants. The image below shows these variants for one
person object under study.
Figure 3.2: Face variants of one person object under study
24
3.4.2.2 Facial recognition model training and validation
After the face data gathering process, the next phase was model training for facial
recognition. Facial recognition refers to the capability to identify a person solely relying on
their facial features (Farayola and Dureja, 2020). The face recognition was carried out
through 5 steps outlined in Figure 3.2 below obtained from the Handbook of Face
Recognition (Wheeler et al, 2011).
Figure 3.3: Face recognition processing flow (Wheeler et al, 2011).
The model was written in Python programming language with OpenCV as the main library.
Python is a high level programming language with very simple syntax (Nitnaware, 2019).
The language was created by Guido van Rossum in 1985. Over the years it has evolved and
has been improved and to date it is considered one of the best languages for machine
learning models. OpenCV is a Python library that provides a real time computer vision
interpretation. It supports execution for machine learning capabilities (OpenCV, 2021).
The deep learning model was developed using Tensorflow. Tensorflow is a framework for
machine learning specifically designed for neural networks (Farayola and Dureja, 2020). The
face detection neural network used was Multi Task Cascaded Convolutional Neural Network
(MTCNN). MTCNN is a neural network which detects faces and facial landmarks on pictures
(Zhang et.al, 2016).
Face
detection
Locating a face that is at a building security checkpoint
Face
alignment
Normalising the face geometry and photometrics to be
consistent with the database
Feature
extraction
Extract features from the face to be used for recognition
purposes
Feauture
Matching
Match the face features against known faces in the databases
Face
Identification
Output the found face and ask the visitor to confirm the details
25
Zhang et al. (2016) proposed MTCNN as a multi-task cascaded CNNs based framework for
joint face alignment and detection. The methods regularly outperform state-of-the-art
techniques on a number of difficult benchmarks for face detection and face alignment while
maintaining real-time performance, according to experimental results. Different academics
and scientists have used the neural network over time.
The model was validated before actual use began. Below shows images of the model
validation to verify one human object under study.
Figure 3.4: Facial recognition made on a clear face
Figure 3.5: Facial recognition made on a face with an object on it
26
Figure 3.6: Facial recognition made on a face with a face mask on chin
3.4.2.3 Data gathering of identity documents
One of the most important stages at the security checkpoints is verifying the identity of an
individual from a valid identity document such as a passport, national identity (ID) card or a
driver’s licence. The researcher proposed that when someone is entering a building for the
first time they should be asked for an identity document and the platform will try to get the
details of the individual from the issued document, hence the pattern recognition part of the
research.
There are 5 variants of identity documents usually accepted at most buildings in Zimbabwe
and these are plastic ID card, metal ID card, A4 paper ID, metal driver’s licence and a valid
passport. The researcher also noted that there is not a publicly available database of these
documents, therefore the researcher created a database of such documents. The researcher
gathered ten (10) of each of the 5 different documents and in the end had 50 different
identity documents. Personal data was not made public as it contains sensitive identity
details of individuals.
The following pictures show Zimbabwean identity documents indicating the necessary parts
(first name, last name and national ID number) of each document that is desired in this
research.
27
Figure 3.7: Zimbabwean Passport
28
Figure 3.8: Zimbabwean Paper Drivers Licence
29
Figure 3.9: Zimbabwean Metal Drivers Licence
30
Figure 3.10: Zimbabwean Plastic ID Card
3.4.2.4 Pattern recognition model training on identity documents
After gathering the identity documents, the next stage was training and validating the
pattern recognition model. Pattern recognition refers to the use of machine learning
techniques and algorithms to identify patterns supplied (Ahmed and Basha, 2021). Labelled
data obtained in the earlier stage was trained on how to identify the patterns on each of the
different documents so that when supplied unlabelled data will recognise patterns of a
visitor's identity details.
The pattern recognition model was implemented using a convolutional neural network. The
researcher made use of Tensorflow and Keras to achieve the desired results.
3.5 Model Development
Model development is an iterative process in which several models are created, tested, and
improved upon until the necessary requirements are met. Three (3) different models were
developed for the purpose of this research.
31
The first one was a facial recognition deep learning model which was used to recognise
human faces from a live camera feed as people pass through a security checkpoint. The
second model was a pattern recognition machine learning model used to identify a person’s
first name, last name and national registration number from their identity document. These
first two models have been described in the research design stage. The last model was a
security checkpoints system simulation model developed to explore and demonstrate how
the overall research proposal would work. The system model was developed using ReactJS,
NodeJS and MongoDB.
React is a JavaScript-based user interface library (see "React - A JavaScript library for
building user interfaces," 2022). Creating interactive user interfaces is a breeze using React.
Create basic views for each stage of your project, and when your data changes, React will
update and render the necessary components. Declarative views make your code more
predictable and easier to debug. You can make discrete components that manage their own
state and then combine them to make complex user interfaces. Because component
functionality is written in JavaScript rather than templates, you can simply convey rich data
through your app while keeping state off the DOM. React can also render on the server with
Node and power mobile apps with React Native.
Node.js runtime is a V8 JavaScript engine ("About | Node.js," 2022). Node.js is a network
application framework that employs an asynchronous event-driven JavaScript engine to
develop scalable network applications. HTTP is recognised as a first-class citizen in Node.js,
and streaming and low latency are prioritised. As a result, Node.js is well-suited to the
creation of web libraries and APIs. The researcher will create an API using NodeJS in this
project.
MongoDB is an open-source, cross-platform document-oriented database programme
("MongoDB: The Application Data Platform", 2022). MongoDB is a NoSQL database that uses
JSON-like documents and optional schemas to store data. MongoDB is a database developed
by MongoDB Inc. and released under the terms of the Server Side Public License (SSPL).
32
MongoDB will be the database management system used. The database will be hosted in the
cloud using Mongo Compass.
Below are some of the simulation screenshots.
Figure 3.11: Security Checkpoint Simulation System Facial Recognition Screen
Figure 3.12: Security Checkpoint Simulation System Initiate Pattern Recognition Screen
Figure 3.13: Security Checkpoint Simulation System Pattern Recognition Screen
33
Figure 3.14: Security Checkpoint Simulation System Event Success Screen
3.6 Conclusion
This chapter went over every facet of how the research questions were answered and how
objectives were met. The researcher proposed an action research with a mixed qualitative
and quantitative research approach. The researcher further laid out the research design and
the steps of the research design mainly covered data gathering of faces, facial recognition
model development, data gathering of identity documents and pattern recognition model
development. In addition, the researcher outlined the three (3) developed models. In the
next chapter, the research discusses the results of this research.
34
Chapter 4 - Results and Discussion
4.1 Introduction
This chapter presents the results obtained from deployment and evaluation of the security
checkpoint simulation model as well as the pattern and facial recognition models. In addition
it presents results of surveys conducted to understand more on access control mechanisms
being used in the world and Zimbabwe. The chapter is organised in a topical manner, with
each topic tabulating results and a discussion on the results.
4.2 Access Control Mechanisms being used in Gweru buildings
Building / Wing
No access control
mechanism
Manual log book
Electronic log book
Bahadur Centre
CAIPF Building
Development House
Electricity House
First Mutual Building
Government Complex
Gweru Town House
MCH Building
MIPF Building
Moonlight
Megawatt Building Wing 1
Megawatt Building Wing 2
Megawatt Building Wing 3
Old Mutual (CABS)
Building
Second Street Shopping
Mall
Standard Chartered
Zimnat Building
52.94%
47.06%
0%
Table 4.1: Access control mechanisms in use in public buildings
Of the 17 public buildings in Gweru, only 47.06% have an access control mechanism at a
security checkpoint. Of these 47.06 public buildings not a single building is making use of an
electronic based visitors logging and auditing system.
35
4.3 Crime records most prevalent on public buildings between April 2021 and March
2022
Crime
Count
Percentage (%)
Vandalism
5
12.5
Theft and burglary
3
7.5
Intrusion
2
5
Assets loss
6
15
Assets and equipment
damage
7
17.5
System disruptions
3
7.5
Data breaches
1
2.5
Offences against persons
which include general
violence and ugly
confrontations
0
0
Unauthorised access to
utilities such as electricity,
telephones and internet
4
10
Disorder and drug affected
behaviour
6
15
Other
3
7.5
Total
40
Table 4.2: Crime records performed on building between April 2021 and March 2022
It is evident from these results that crimes are a frequent thing on public buildings in Gweru.
Of these crimes, the top three (3) crimes are assets and equipment damage, assets loss and
disorder and drug affected behaviour. Hence there is a need to have an access control
36
mechanism necessary to log and audit people into and out of a building for preventive
purposes.
4.4 Facial Recognition Model Accuracy and Time Taken Results
Human subject
Accuracy Score (%)
Time Taken (s)
Dzinaishe
95.03
8
Beverly
93.00
10
Benedict
91.71
11
Faith
91.34
9
Tinashe
90.01
12
Table 4.3: Results from human objects studied
Five (5) human subjects were studied to observe accuracy in matching with existing faces as
well as the time it would take the model to process and output a result. The highest result
was 95.03 and the lowest score 90.01. These are reasonable times in which the model can
run in and produce results. Observing visitors by facial recognition as they come into a
building is therefore an efficient way of improving service delivery in public buildings.
4.5 Pattern Recognition Model Accuracy Results
Human
Object
Plastic ID
Accuracy
(%)
Metal ID
Accuracy
(%)
Paper ID
Accuracy
(%)
Passport
Accuracy
(%)
Metal
Drivers
Licence
Accuracy
(%)
Paper
Drivers
Licence
Accuracy
(%)
Dzinaishe
Mpini
80
-
-
11
0
0
Benedict
Chadiwa
85
-
-
0
0
0
Winfilder
Mpini
-
64
-
0
-
-
37
Constance
Mpini
82
-
0
0
-
-
Table 4.4: Results on accuracy of pattern recognition model on identifying identity documents
patterns
The pattern recognition model performance was really poor on the majority of commonly
accepted identity documents in Zimbabwe. The best performing type of document was the
plastic national ID with accuracy ranging from 80 to 85 for 3 different people. It is
recommended to make use of this document more compared to all the other documents on
the security checkpoints. The poor performance on paper documents is mainly due to
different handwritings on the paper documents. The poor performance on the metal ID card,
licence and passport is probably due to the different fonts and sizes for text on each of the
documents.
4.6 Conclusion
It was noted in this research that buildings in Zimbabwe have poor access control
mechanisms yet crimes on buildings do happen frequently. Adoption of an access control
mechanism is therefore important to prevent these crimes and be able to deal with them
when they occur. The research further proved the feasibility of merging pattern and facial
recognition techniques in improving security checkpoints, service delivery and privacy of
visitors.
38
Chapter 5 - Recommendations and Conclusion
5.1 Introduction
A simulation model to demonstrate the feasibility of merging pattern and facial recognition
to improve service delivery and privacy concerns at a security checkpoint was developed in
this research. In this chapter, the last chapter of this researcher, the researcher will outline
the aims and objectives realisation of the research, challenges faced whilst carrying out the
project and finally conclude with recommendations for future work.
5.2 Aims and Objectives Realisation
The aim of this study was to develop a security checkpoint assistant model armed with
computer vision and pattern recognition techniques so as to quickly serve customers in a
safe way whilst at the same enforcing more efficient building security with the following
objectives:
To streamline a public building visitors checkin and checkout process through facial
recognition.
To register first time customers through pattern recognition done on national
identity documents, passports or driver’s licence.
To screen visitors entering a building against a company’s security watch list thereby
allowing selective access to a building.
To provide insights and analytics on building visitors activity.
To pre register guests in the case of many visitors who will visit the building at one
time.
All the objectives of this research were met through developed simulation models.
5.3 Challenges Faced
The most difficult problem for the students was acquiring access to data in separate
buildings. When dealing with sensitive and confidential material, institutions prefer to
provide aggregate level data. Individual level data is frequently essential for doing rigorous
empirical research, although aggregate level data might be beneficial for obtaining a "broad
stroke" picture of a certain institution's students. Individual level data, or information and
39
data at the student level, is more difficult to come by, yet it's necessary when analysing
specific groups and student outcomes.
The majority of the data included in the study was self-reported. As with any research effort
involving human attitudes and behaviours, relying on self-reported data means that our data
is only as good as the responses we receive. The veracity of self-reported data is unknown
without the availability of data for cross-checking, which is a barrier in student population
research.
The researcher had a difficult time raising enough money to self-fund the research effort.
Given the current economic climate, monetary incentives are unlikely, or incentivization
leads to limited participation. As a result, researchers are forced to think of new approaches
to stimulate participation while working with limited resources.
Furthermore, the researcher had hoped to build an actual robot at the outset of the project,
but due to funding constraints, he was unable to do so and instead chose to use a simulation
model.
Finally, most research is done in groups, but distance learning has no such culture. That
instance, in the realm of distance education, good research typically involves a team effort.
Many firms, on the other hand, do not encourage team-based efforts, as indicated by the lack
of technologies that allow multiple people to engage and cooperate on the same project.
5.4 Recommendations for future work
The developed model does not address the “way finding problem”. This is a scenario in which
someone enters a building and is not sure where they are going, hence they would need to
be given directions to the office or specific spot in the building which they should go to. In
the meantime, a human being should be readily available working together with the system
for assisting scenarios like this. It would be best to develop a system which has a solution to
this problem.
The research may be extended to make use of voice recognition, which is a great biometric
technique to identify a human being. Voice recognition is way quicker than facial recognition
40
as the communication between device and the human being is more efficient. With voice
recognition, the “way finding problem” can be quickly addressed as the device can make a
conversation with a human being.
Nowadays, because of Covid 19, people are required to move around with face masks so as
to reduce the chances of spreading the virus. The proposal relies on facial recognition which
works best when someone does not have a face mask. The facial recognition model can only
identify a person if the face mask is on the chin and cannot detect a face if a face mask is
covering the mount and or nose. It will be best to be able to identify people even when they
have a face mask.
On pattern recognition done on identity documents, the pattern recognition model managed
to work efficiently with the plastic national identity document. It will be best to train the
model to be able to identify patterns on other identity documents which include the plastic
ID card, paper ID card, drivers licence and passport.
41
References
About | Node.js. Node.js. (2022). Retrieved 12 March 2022, from
https://nodejs.org/en/about/.
Ahmed, S. and Basha, M., 2021. Pattern recognition, an introduction.
Alodainy, A., 2022. Fingerprint, Security level, and possible improvements. [online] Available
at:
<https://www.researchgate.net/publication/360973126_Fingerprint_Security_level_and_p
ossible_improvements> [Accessed 4 May 2022].
ARAKALA, A., 2009. SECURE AND PRIVATE FINGERPRINT-BASED AUTHENTICATION.
Bulletin of the Australian Mathematical Society, [online] 80(2), pp.347-349. Available at:
<http://researchgate.net/publication/231781930_Secure_and_private_fingerprint-
based_authentication> [Accessed 4 May 2022].
Arulogun, O., Omidiora, E., Olaniyi, O. and Ipadeolaa, A., 2008. Development of Security
System using Facial Recognition. [online] Available at:
<https://www.researchgate.net/publication/235934529_Development_of_Security_Syste
m_using_Facial_Recognition> [Accessed 27 January 2022].
Bhalla, V. and Gupta, M., 2021. A Paper on Fingerprint Recognition. Journal of Pharmaceutical
Research International, [online] pp.3429-3436. Available at:
<https://www.researchgate.net/publication/358441265_A_Paper_on_Finger_Print_Recogn
ition> [Accessed 4 May 2022].
Bhandari, A., Wattamwar, K., Preeti, P., Bhatt, S. and Gite, B., 2010. Automated Visitors
Management System. Interscience Management Review, pp.102-104.
42
Castelblanco, A., Solano, J., Lopez, C., Rivera, E., Tengana, L. and Ochoa, M., 2020. Machine
Learning Techniques for Identity Document Verification in Uncontrolled Environments: A
Case Study.
Bryman, A. and Bell, E, 2011. Business Research Methods. 3rd edition, Oxford University
Press
Casey, D., 2008. Turning log files into a security asset. Network Security, 2008(2), pp.4-7.
Chaudhary, B., 2021. Visitor Management System. International Journal for Research in
Applied Science and Engineering Technology, [online] 9(5), pp.750-751. Available at:
<http://dx.doi.org/10.22214/ijraset.2021.34279> [Accessed 10 September 2021].
Chung, T., 2018. A Study on the Development of Security Check Standards for Government
Office Buildings. Journal of the Korean Society of Hazard Mitigation, [online] 18(5). Available
at:
<https://www.researchgate.net/publication/330630636_A_Study_on_the_Development_of
_Security_Check_Standards_for_Government_Office_Buildings> [Accessed 10 September
2021].
Clancy, T., Kiyavash, N. and Lin, D., 2003. Secure smartcardbased fingerprint authentication.
Proceedings of the 2003 ACM SIGMM workshop on Biometrics methods and applications -
WBMA '03, [online] Available at:
<https://www.researchgate.net/publication/234828386_Secure_smartcardbased_fingerpr
int_authentication> [Accessed 4 May 2022].
Cobalt Robotics. 2022. Physical and cyber security services - Cobalt Robotics. [online]
Available at: <https://www.cobaltrobotics.com/service/security/> [Accessed 27 January
2022].
43
Congressional Research Service, 2017. Federal Building and Facility Security: Frequently
Asked Questions.
Creswell,J.W., 2013. Research design: Qualitative, quantitative, and mixed methods
approaches. Sage publications.
Face-rec.org. 2021. Face Recognition Homepage - Databases. [online] Available at:
<https://www.face-rec.org/databases/> [Accessed 15 September 2021].
Farayola, M. and Dureja, A., 2020. A Proposed Framework: Face Recognition With Deep
Learning. INTERNATIONAL JOURNAL OF SCIENTIFIC & TECHNOLOGY RESEARCH, [online]
9(7). Available at: <https://www.ijstr.org/final-print/jul2020/A-Proposed-Framework-
Face-Recognition-With-Deep-Learning.pdf> [Accessed 10 September 2021].
Forte, D., 2005. Log management for effective incident response. Network Security, 2005(9),
pp.4-7.
Forte, D., 2009. The importance of log files in security incident prevention. Network Security,
2009(7), pp.18-20.
Gautam, K., Sharma, N., Kumar, P. and Mishra, V., 2021. COVID 19 Visitor Management
System. In: 2021 International Conference on Computational Intelligence and Knowledge
Economy (ICCIKE).
Gao, Q., 2022. A secure and private fingerprint-based authentication system. [online]
Available at:
<https://www.researchgate.net/publication/316918067_A_secure_and_private_fingerprin
t-based_authentication_system> [Accessed 4 May 2022].
Ghaithi, H. and Eaganathan, U., 2016. A BRIEF STUDY AND IMPLEMENTATION OF VISITOR
MANAGEMENT SYSTEM FOR ASIA PACIFIC UNIVERSITY, MALAYSIA. International Journal
44
of Advanced Research in Science Engineering, [online] 5(4). Available at:
<http://ijarse.com/images/fullpdf/1460385237_531V.pdf> [Accessed 10 September
2021].
Guennouni, S., Mansouri, A. and Ahaitouf, A., 2020. Biometric Systems and Their
Applications. Visual Impairment and Blindness - What We Know and What We Have to Know,
[online] Available at: <https://www.intechopen.com/chapters/65920> [Accessed 4 May
2022].
Isa, M., Yahaya, Y., Halip, M., Khairuddin, M. and Maskat, K., 2010. The design of fingerprint
biometric authentication on smart card for PULAPOT main entrance system. 2010
International Symposium on Information Technology, [online] Available at:
<http://10.1109/ITSIM.2010.5561969> [Accessed 10 September 2021].
Karimah, N., Noor, M., Sulaiman, J. and Chen, K., 2007. Development of Visitor Management
System Using Smart Card: UMP Case Study. [online] Available at:
<http://umpir.ump.edu.my/id/eprint/976/> [Accessed 10 September 2021].
Kapoor, M., 2018. Security Assessment case studies of public buildings in India. International
Journal of Safety and Security Eng, 8(2).
Knightscope. 2022. Knightscope. [online] Available at: <https://www.knightscope.com/>
[Accessed 27 January 2022].
Kosar, J. and Ahmed, F., 2000. Building Security into Schools. Schools. [online] Available at:
<https://www.researchgate.net/publication/234619763_Building_Security_into_Schools/
> [Accessed 10 September 2021].
Gautam, K., Sharma, N., Kumar, P. and Mishra, V., 2021. COVID 19 Visitor Management
System. 2021 International Conference on Computational Intelligence and Knowledge
Economy (ICCIKE), [online] Available at:
45
<https://www.researchgate.net/publication/351155185_COVID_19_Visitor_Management_
System> [Accessed 10 September 2021].
Linklater, G., 2016. Multi Stage Physical Access Control. BSc. Rhodes University.
Mawby, R., 2014. Commercial Burglary. In: G. Martin, ed., The Handbook of Security. [online]
Available at: <https://link.springer.com/book/10.1007/978-1-349-67284-4> [Accessed 27
April 2022].
MIT, 2022. Security Controls. [online] Web.mit.edu. Available at: <https://web.mit.edu/rhel-
doc/4/RH-DOCS/rhel-sg-en-4/s1-sgs-ov-controls.html> [Accessed 21 April 2022].
Mohsienuddin, S. and Sabri, M., 2020. FACIAL RECOGNITION TECHNOLOGY. SSRN Electric
Journal, [online] 7(6). Available at:
<https://www.researchgate.net/publication/354790385_FACIAL_RECOGNITION_TECHNO
LOGY> [Accessed 4 May 2022].
MongoDB: The Application Data Platform. MongoDB. (2022). Retrieved 12 March 2022, from
https://www.mongodb.com/.
Naiguo, W., Xiangwei, Z. and Jian, Z., 2015. Research of ID Card Recognition Algorithm Based
on Neural Network Pattern Recognition. In: International Conference on Mechatronics,
Electronic, Industrial and Control Engineering (MEIC 2015).
Nitnaware, R., 2019. Basic Fundamental of Python Programming Language and The Bright
Future. [online] Available at: <https://www.researchgate.net/publication/350192013>
[Accessed 10 September 2021].
Okamura, M. and Norifusa, M., 2014. Method and system for providing terminal security
checking service. [online] Available at:
46
<https://www.researchgate.net/publication/302766972_Method_and_system_for_providi
ng_terminal_security_checking_service/citations> [Accessed 10 September 2021].
Oktaviandri, M. and Keat, F., 2018. Design and Development of Visitor Management
System. Journal of Intelligent Manufacturing & Mechatronics, [online] 1(1). Available at:
<https://core.ac.uk/download/pdf/211030682.pdf> [Accessed 27 January 2022].
OpenCV. 2021. Home - OpenCV. [online] Available at: <https://opencv.org> [Accessed 15
September 2021].
Phin, P., Abbas, H. and Kamaruddin, N., 2020. Physical Security Problems in Local
Governments: A Survey. Journal of Environmental Treatment Techniques, [online] 8(2),
pp.679-686. Available at: <http://jett.dormaj.com> [Accessed 27 April 2022].
Queensland Government Department of Housing and Public Works, 2017. Security
Management of Government Buildings. Brisbane.
React - A JavaScript library for building user interfaces. (2022). Retrieved 12 March 2022,
from https://reactjs.org/
Rivera, J., 2010. VMS Support: A mobile-based support to computerized visitor management
system. Software Impacts, 8, p.100056.
Ryan, M. and Hanafiah, N., 2015. An Examination of Character Recognition on ID card using
Template Matching Approach. In: International Conference on Computer Science and
Computational Intelligence (ICCSCI 2015).
Safeguard Security Zimbabwe. 2022. Safeguard Crime Report March 2021 | Safeguard
Security Zimbabwe. [online] Available at: <https://www.safeguard.co.zw/safeguard-crime-
report-march-2021/> [Accessed 4 May 2022].
47
Sandhu, R. and Samarati, P., 1994. Access Control: Principles and Practise. IEEE
Communications Magazine, pp.40-48.
Satari, B., Abd Rahman, N. and Zainal Abidin, Z., 2014. Face recognition for security efficiency
in managing and monitoring visitors of an organization. 2014 International Symposium on
Biometrics and Security Technologies (ISBAST), [online] Available at:
<https://www.researchgate.net/publication/282187833_Face_recognition_for_security_eff
iciency_in_managing_and_monitoring_visitors_of_an_organization> [Accessed 10 September
2021].
Sinha, A., Nguyen, T., Kar, D., Brown, M., Tambe, M. and Jiang, A., 2015. From physical security
to cybersecurity. Journal of Cybersecurity, p.tyv007.
SMP Robotics - Autonomous mobile robot. 2022. Security robots outdoor patrolling, video &
thermal surveillance for security service companies. [online] Available at:
<https://smprobotics.com/security_robot/> [Accessed 27 January 2022].
Song, Z., Nguyen, K., Cho, C. and Gao, J., 2021. Camera-Based Security Check for Face Mask
Detection Using Deep Learning. [online] Available at:
<https://www.researchgate.net/publication/353511105_Camera-
Based_Security_Check_for_Face_Mask_Detection_Using_Deep_Learning/> [Accessed 10
September 2021].
Wan, M., Chen, Z. and Guo, J., 2021. Optimization of security check efficiency in subway
station based on Anylogic: A case study of Nanchang Metro. Journal of Intelligent & Fuzzy
Systems, [online] pp.1-9. Available at:
<https://www.researchgate.net/publication/351669271_Optimization_of_security_check_
efficiency_in_subway_station_based_on_Anylogic_A_case_study_of_Nanchang_Metro/citatio
ns> [Accessed 10 September 2021].
United Nations, 2021. Technology and Innovation Report 2021.
48
Usmani, Z., Irum, S., Afzal, H. and Mahmud, S., 2013. How to Build an Automated Fingerprint
Identification System. 2013 International Symposium on Biometrics and Security
Technologies, [online] Available at:
<https://www.researchgate.net/publication/260739557_How_to_Build_an_Automated_Fin
gerprint_Identification_System> [Accessed 4 May 2022].
Victorian Auditor - General's Office, 2019. Security of Government Buildings. Victorian
Government Printer.
Wheeler, F., Liu, X. and Tu, P., 2011. Handbook of Face Recognition. 2nd ed.
Zhang, B., 2015. Intelligent Building Security System Design based on Internet of Things
Technology. In: 4th National Conference on Electrical, Electronics and Computer
Engineering.
Zhang, K., Zhang, Z., Li, Z. and Qiao, Y., 2016. Joint Face Detection and Alignment Using
Multitask Cascaded Convolutional Networks. IEEE Signal Processing Letters, [online] 23(10),
pp.1499-1503. Available at: <https://arxiv.org/pdf/1604.02878.pdf>.
49
Appendices
Appendix A: Manual log sheet
50
Appendix B: Access control mechanisms in different public buildings in Gweru
Observations Score Sheet
Name of Observer …………………………………………………………..
Date of Observation …………………………………………………………..
Time of Observation …………………………………………………………..
Place of Observation ………………………………………………………….
Object being Observed ………………………………………………………….
Building / Wing
No access control
mechanism
Manual log book
Electronic log
book
Bahadur Centre
CAIPF Building
Development House
Electricity House
First Mutual
Building
Government
Complex
Gweru Town House
MCH Building
MIPF Building
Moonlight
Megawatt Building
Wing 1
Megawatt Building
Wing 2
Megawatt Building
Wing 3
Old Mutual (CABS)
Building
Second Street
Shopping Mall
Standard Chartered
Zimnat Building
51
Appendix C: Interview Guideline for Public Buildings
Name of Interviewer …………………………………………………………..
Date of Interview …………………………………………………………..
Time of Interview …………………………………………………………..
Place of Interview ………………………………………………………….
On a scale of 1-10 how high is security a concern in your building?
………………………………………………………………………………………………………
What security control mechanisms have you implemented in your building?
………………………………………………………………………………………………………
How much is your annual security budget?
………………………………………………………………………………………………………
How many times have you encountered the following crimes in your building?
Crime
Count
Vandalism
Theft and burglary
Intrusion
Assets loss
Assets and equipment damage
System disruptions
Data breaches
Offences against persons which include general violence and ugly
confrontations
Unauthorised access to utilities such as electricity, telephones and
internet
52
Disorder and drug affected behaviour
Other
I am currently working on developing a security checkpoint assistant robot armed with
computer vision and pattern recognition techniques so as to quickly serve customers in a
safe way whilst at the same enforcing more efficient building security.
Would you be interested in such a platform?
…………………………………………………………………………………………………….
Other comments you have
…………………………………………………………………………………………………………………………………..……
……………………………………………………………………………………………………………………………..…………
………………………………………………………………………………………………………………………..………………
…………………………………………………………………………………………………………………..……………………
……………………………………………………………………………………………………………..…………………………
………………………………………………………………………………………………………..………………………………
…………………………………………………………………………………………………….
ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
Biometric identification has become an important, or it can be said, the most important part of security in the world, these biometric identifications consist of many biometric securities, like a fingerprint, face unlock, iris, and so many other biometrics. Each one of these biometrics has issues and level of security that can be used for. This paper will focus on the fingerprint biometric issues including some experiments and self-experiences. In the first part of this paper, I will discuss the fingerprint, its level of security and, how far we can trust this biometric in security. For this purpose, we will have at look how it works, how it differs from person to person, and how to optimize the checking control. In the next part I will discuss the possibility to copy the fingerprint t, using different methods, and different materials, also I will discuss how easy it is to copy the fingerprint, meanwhile I will propose some solutions to identify the copied one from the original one if possible. Finally, all the drawbacks will be collected and a solution or a way to improve the security level will be provided.
Book
Full-text available
This book covers the primary and supportive topics on pattern recognition with respect to beginners understand-ability. The aspects of pattern recognition is value added with an introductory of machine learning terminologies. This book covers the aspects of pattern validation, recognition, computation and processing. The initial aspects such as data representation and feature extraction is reported with supportive topics such as computational algorithms and decision trees.
Conference Paper
Full-text available
According to the World Health Organization (WHO), wearing a face mask is one of the most effective protections from airborne infectious diseases such as COVID-19. Since the spread of COVID-19, infected countries have been enforcing strict mask regulation for indoor businesses and public spaces. While wearing a mask is a requirement, the position and type of the mask should also be considered in order to increase the effectiveness of face masks, especially at specific public locations. However, this makes it difficult for conventional facial recognition technology to identify individuals for security checks. To solve this problem, Deep Learning algorithms are proposed to develop a comprehensive mask detection through a camera. CNN, AlexNet, VGG16, and Facial Recognition Pipeline with FaceNet are the Deep Learning algorithms that are used to classify the features in each scenario. This project uses multiple open datasets, Google Image Search, and simulated photos. We aim to build a comprehensive camera system that is able to cover four major issues:
Article
Full-text available
Now a day to day lot of programming languages that can realize our needs, but the most important question is how to teach programming to beginner students. In this paper we suggest using Python for this purpose, because it is a programming language that has neatly organized syntax and powerful tools to solve any task. Moreover it is very close to simple math thinking. Python is chosen as a primary programming language for freshmen in most of leading universities. Firstly, this paper proposes advantages of Python language in relation to C++ and JAVA. Python is a suitable language for both learning and real world programming. Python is a powerful high level, object-oriented programming language created by Guido van Rossum. In this paper we first introduce you to the python programming characteristics and features. This paper also discusses about the reasons behind python being credited as the most fastest growing programming language in the recent times supported by research done over the articles procured from various magazines and popular websites. This paper features about the characteristics and most important features of python language, the types of programming supported by python and its users and its applications. Finally, Python has a bright future ahead for developing almost every kind of applications whether it is web applications or game applications.
Article
Full-text available
VMS Support is a mobile-based application which is intended to integrate with a computerized visitor management system (CVMS). The application is used as a receiver of visit notification, respond to the CVMS and update the current whereabout. Visit notification was built using a mobile offline push notification which will not force the organization to invest in any internet provider and subscribe to a premium push notification service. User may respond to visit notification by sending a message that may be visible to the CVMS, and update its current whereabout to guide the visitor correctly.
Article
Full-text available
Physical security refers to the control of access into organizations, buildings, rooms, and information technology (IT) peripherals. However, physical security may be overlooked by organizations because they are more concerned about information security; this is because the organization assumes that those granted access can be trusted. The physical security is not a new issue in a local government environment; however, in most cases, hackers are to blame, while the actual culprits may be the employee(s) of the local government itself. This paper is a survey done to investigate the problems faced by a local government and the measures needed to be taken to keep their physical access secure. The subject of the research is chosen from among municipal councils in Malaysia since they hold various private information about the residents of the area, where physical security awareness is still low among the members of the organization. As a case study, the Kota Bharu Municipal Council (KBMC) was selected and its security problems were identified through a research comprising a mixed method of quantitative (questionnaire and observation) and qualitative (interview) techniques. The respondents of the research were eight employees of the IT Department, while the solution to their security problems was derived through interviewing its IT Officer. The researchers also discuss if KBMC is able to apply other local government's solutions to their own security problems. The discussion reveals that the security awareness program is the most suitable solution to the KBMC's security problems since it enhances security awareness of top management officers and enables the employees to be aware of their responsibilities in their daily work routine.
Article
Rational and orderly passenger organization in subway station can improve travel efficiency and safety. How to improve the efficiency of security check, which has a significant impact on the passenger organization, to reduce passengers staying time in the terminal level has the potential to improve service satisfaction. The Weidong Station of Line 1 of Nanchang Metro was selected as the research object. Anylogic software was used to simulate the process of arrival, ticket purchase, security check, gate-crossing machine etc. The results indicate that the dynamic adjustment of security checkpoints and fast passage for passengers without luggage can improve the security efficiency and relieve the congestion in the station hall.
Article
This paper present the Automated Visitors Management System A visitor management system tracks the usage of a public building or site. By gathering increasing amounts of information, a visitor management system can record the usage of the facilities by specific visitors and provide documentation of visitor’s whereabouts. Because a visitor management system provides a record of building use, these systems are frequently used to complement building security systems and access control systems. As electronic visitor management systems become more common and more powerful, these systems are taking over many of the functions of building security and access control.
Chapter
Distributed (i.e. mobile) enrollment to services such as banking is gaining popularity. In such processes, users are often asked to provide proof of identity by taking a picture of an ID. For this to work securely, it is critical to automatically check basic document features, perform text recognition, among others. Furthermore, challenging contexts might arise, such as various backgrounds, diverse light quality, angles, perspectives, etc. In this paper we present a machine-learning based pipeline to process pictures of documents in such scenarios, that relies on various analysis modules and visual features for verification of document type and legitimacy. We evaluate our approach using identity documents from the Republic of Colombia. As a result, our machine learning background detection method achieved an accuracy of 98.4%, and our authenticity classifier an accuracy of 97.7% and an F1-score of 0.974.