FINAL YEAR PROJECT LITERATURE REVIEW

1. A STUDY ON ACCESS CONTROL SYSTEM USING FACE

RECOGNITION (Gabit Tolendiyev, Byung Gook Lee, 2019).

This project presents the use of facial recognition technology as a method of

access control. This is because facial recognition has shown great performance

compared to other biometric technology, making it very popular.

Facial recognition has been a controversial topic in computer vision and

pattern recognition in the last few decades. Facial recognition stands as one of the

most popular biometric system in existence today, greatly used in unlocking phones,

doors, verifying payments and also at entrances of large-scale event instead of ID

cards (Okumura & Hoshino et. al., 2017).

Fingerprint, palm vein and eye iris recognition are other biometric method

mentioned in this project but facial recognition was most preferred because it does not

require any physical contact, non-invasive and secure. Below, a table is given,

comparing a few biometric systems in terms of technical and non-technical

weaknesses (Gabit & Byung, 2019).

facial recognition:

- Illumination: low illumination and direct light effect on image acquiring;

- Posture: persons do not directly look at the camera;

- Occlusion: partial occlusion of face by other objects, e.g. hands;

- Accessories: sunglasses, cap, etc.

In order to address these problems and achieve better recognition accuracy

many approaches have been proposed (Okumura & Hoshino et. al., 2017; Schroff &

Kalenichenko et. al., 2015; Deng & Guo et. al., 2018; Amos & Ludwiczuk et. al.,

2016):

- Employing large-scale Deep CNN containing more than 1.6B parameters.

- Training the model on very huge dataset, e.g. more than 200M image of 8M

different identities (Schroff & Kalenichenko et. al., 2015);

- Loss function that distinguish two face images (Contrastive loss, triplet loss, center

loss, large margin softmax, angular softmax, and large margin cosine loss, and cosine

loss).
An illustration of how facial recognition is implemented in access control systems is

shown below:

(Gabit & Byung, 2019)

Working Principles

The system developed was built on open-source face recognition model. It

features a camera which collect a snapshot of the user’s face, crops it and compares it

with templates stored on a database. The comparison is calculated using Euclidean

Distance which computes the similarity between the facial dataset and a numerical

value. If the numerical value is greater than a threshold, the datasets are regarded as

dissimilar otherwise, similar. The system also keeps log of entry and exit time of users

and can be used as an attendant management system as well.

Conclusion

This project makes use of facial recognition for access control. The

researchers believe future works will be based on improving the system to

automatically register a new face and recognize the face on next appearance.
2. IOT BASED DOOR ACCESS CONTROL USING FACE

RECOGNITION (Harsha H Prasad, 2019).

This project starts by acknowledging the traditional security methods which

have been used to enforce privacy. This traditional security methods which includes

the use of keys, identification card or passwords have been used to restrict access to

certain resources or information but regarding their flaws, they became less popular.

This literature proposed an access control system using face detection and

recognition. The system consists of image capture, face detection and recognition,

email notification and automatic door access management. OpenCV was used in the

facial recognition system because it is able to compress facial templates without

loosing vital facial properties. The facial templates can be stored in a database to be

retrieved afterwards.

The system also features a GSM module which allows the door to be

controlled remotely and even facilitates sending of email to security administrator.

The Pi Camera used captures images to be used by the facial recognition system.

The project also states major areas of commercial interest for the development

of facial recognition technology, which includes bio science, law enforcement and

surveillance, human computer interaction, multimedia management, passport check,

criminal investigations, access control management. This high demand has invented

cheaper and flexible designs for facial recognition, making it a popular choice.

Also, like any other technology, facial recognition has some difficulties in its

use such as angle or orientation of face, lighting, and facial features like beards or

even glasses. This made the researcher choose OpenCV based facial recognition

system using Haar’s face classifier. The micro-controller used here is the Raspberry
Pi micro-controller. A PIR Sensor was also used to detect human based on their heat

radiation.

Problem Identification And Objectives

In thus project, the following problems were highlighted as issues associated

with facial recognition technology:

- Pose variation

- Occlusion

- Expression

- Aging

- Transformation

- Illumination

Some of the objectives of facial recognition systems are as follows:

1. To design real time face detection system.

2. To utilize the face detection system based on Haar classifier.

3. To develop face detection system using open CV.

4. Users could operate on a touch screen to select entering the house by recognizing

face or entering password. For face recognition, an image will be captured by a Pi

camera.
Block Diagram

The block diagram of the proposed IoT based door access control system

using facial recognition is given below:

(Harsha, 2019).

Conclusion

This project was programmed with Python programming language. The

developed system’s efficiency was calculated in terms of facial recognition rate and

this analysis revealed that the system has high performance efficiency.

3. RASPBERRY PI-BASED ACCESS CONTROL USING FACE

RECOGNITION (Ondrej KAINZ, Ján DROZD, Miroslav

MICHALKO, František JAKAB, 2019)

This literature starts by referring to a concept where digital devices collect,

process and analyze data from the surrounding and communicates with one another

via the internet. This concept is termed Internet of Things (IoT). It also mentions that

facial recognition is getting a large attention as it can be used for accelerated

identification. It can be used by banks or insurance companies to identify certain

age of the person passing by. The areas of application of facial recognition technology

is just too broad.

This aim of this project is to develop an access control system which detects

faces and processes those faces for recognition. The result of these recognitions

determine is access will be granted to the user or not.

Methods For Face Detection And Recognition

This project discusses two methods of face detection and recognition:

Histogram of Gradients (HOG) and Haar Classifiers.

Histogram of Gradients divides an image into smaller parts which make it

easier to extract vital information from the image and disregarding the unnecessary.

Histogram of direction of gradients are used to define important features of the image.

A research conducted by Hafiz et al. (2018) utilized HOG and CNN (Convolutional

Neural Networks) and an accuracy of 89% was recorded.

Haar Classifiers is a face detection and recognition method proposed by Paul

Viola and Michael Jones in a literature they published. Haar Classifiers use the

contrast between adjacent pixels instead of pixel intensity. A test was carried out with

Haar Classifiers and other weak classifiers and an accuracy of approximately 80%

was recorded.

Other methods for face detection and recognition mentioned are: Support

Vector Machine and Neural Networks.

Proposal Of Face Recognition System

camera. The camera captures a face and the Raspberry Pi micro-controller analyzes

the face and calculated the Euclidean distance between the stored face templates and

the face before the camera. Face detection was executed using an HOG algorithm.

Based on the result of the Euclidean distance between the two face templates,

recognition is acknowledged or dis-acknowledged and the door locking and unlocking

is demonstrated by a flash of LED bulbs.

The result of test made using the developed system is given in the table below:

(Ondrej & Jan et al., 2019)

Conclusion

The system presented an high detection rate of 90.61% facial recognition rate

of 95.83%. Also, positive results have been realized from different distance from the

camera, with different angle, lighting, and multiple test subjects at the same time.
4. AN INTELLIGENT AUTOMATED DOOR CONTROL SYSTEM

BASED ON A SMART CAMERA (Jie-Ci Yang, Chin-Lun Lai, Hsin-

Teng Sheu and Jiann-Jone Chen, 2013).

This project starts by discussing the Automatic Entrance/Exit Door, stating its

use in businesses, transportation systems to mention a few, especially places where

manual door opening and closing needs to be removed. It mentions the existing

sensor-based automatic door control technologies which includes infrared,

ultrasonic/radio, or other wireless sensing methods. Infrared sensors are subdivided

into active and passive sensors. In active sensors, the sensor radiates a signal and

based on the signal reflected back to it, it can determine position of objects and even

their respective speeds. Passive sensors don’t radiate any signal but have the

capability to receive signal that bounce off object and calculate their position and

even speed from the received signal. Passive sensor are cheaper and hence better

preferred, compared to active sensors.

The researchers noted that although, the sensor-based automatic doors are able

to detect object, they are not capable of predicting the intention of users, giving rise

the frequent false actions, which may be annoying and wastes power as stated by the

researchers.

In this literature, a door access control system was developed to work based on

identification of a human being and corresponding trajectory indicating that the

person wants to pass through the door. Also, an infrared component has been

embedded to prevent the door shutting while someone is passing through.

The design of the door access control system is based on the human detection

and trajectory analysis. It involves face detection, trajectory tracking and statistical

analysis for intention estimation. Below is the flowchart of the control procedures:

(Yang & Lai et al., 2013)

The operation of the system is divided into 3 processes: state transition, human

detection, intention analysis.

State transition process involves the opening and closing of the door. The door

opening action depends on the detection of a human face and probability of the human

walking through the door which is calculated using trajectory analysis. The door

closing action uses infrared to detect any object passing through the door and close

the door only when the doorway is clear. The state transition diagram is given below:
 State Transition Diagram (Yang & Lai et al., 2013).

In this project face and contour detection are used in correctly identifying

human. The face detection is used primarily but the contour detection of the head and

shoulder is used as an secondary option if privacy has to be duly considered (Zeng,

2010).

Intention analysis makes use of trajectory tracking and probability of a user

accessing the door the predict is the user indeed wishes to go through the door or is

just passing by. The face of the user is constantly tracked as he/she approaches the

door using Spatial-Temporal Property.

System Performance Evaluation

Detection and false rate are used the measure the performance of the system.

Face detection used the Viola method (Viola & Jones, 2004), by creating a 10-layer

cascade boosting structure. The face detection rate is pegged at 90% while the false

rate (FR) is calculated as the sum of false acceptance rate (FAR) and false rejection

rate (FRR). The false rate is pegged at less than two in one million times.
Conclusion and Future Works

It is noted that the system has advantages of low false rate (near 0%), high

correct activating rate (99.6%), and short response time (within 2s) from detecting the

target, confirming his intention, to activate the door opening. Moreover, via statistical

analysis on detected face in consecutive time sequence, case of passing persons with

missing face can still be confirmed within 4 s thus to activate the door correctly.

The developed system can also be used as a watch dog since it has a higher CP

rate than existing system. The researchers also noted that the developed system can

also be used as an emergency system for fire detection, only needing a replacement

for a faster DSP micro-controller module. The authors of this literature declared no

conflict of interest.
REFERENCES

- A. Okumura, T. Hoshino, S. Hada, Yugo Nishiyama and M. Tabuchi, “Identity

Verification of Ticket Holders at Large-scale Events Using

Face Recognition,” Journal of Information Processing Vol.25 448-458, June 2017.

- B. Amos, B. Ludwiczuk, M. Satyanarayanan, “OpenFace: A general-purpose face

recognition library with mobile applications,” Technical

report, CMU-CS-16-118, CMU School of Computer Science, 2016.

- F. Schroff, D. Kalenichenko, and J. Philbin, “Facenet: A unified embedding for face

recognition and clustering,” arXiv preprint arXiv:1503.03832, 2015.

- J. Deng, J. Guo, and S. Zafeiriou, “ArcFace: Additive angular margin loss for deep

face recognition,” arXiv:1801.07698, 2018.

- Hafiz, A., Ishraq, A., Manirul, I.: "HOGCNN Based Real Time Face Recognition",

International Conference on Advancement in Electrical and Electronic Engineering

(ICAEEE), 2018.
- Viola, P., Jones, M.: "Rapid Object Detection Using a Boosted Cascade of Simple

Features", Proceedings of IEEE Computer Society Conference on Computer Vision

and Pattern Recognition. CVPR, 2001.

- Zeng, C. Robust Head-Shoulder Detection by PCA-Based Multilevel HOG-LBP

Detector for People Counting. In Proceedings of the IEEE International Conference

on Pattern Recognition, Beijing, China, 23–26 August 2010; pp. 2069–2072.

Viola, P.; Jones, M.J.: Robust real-time face detection. Int. J. Comput. Vis. 2004, 57,

137–154