You are on page 1of 23

TITLE:

DESIGN AND IMPLEMENTATION OF AN


INTRUDER DETECTION SYSTEM WITH
ALERT USING TRANSFER LEARNING AND
MULTIPLE IMAGING SYSTEMS

Name : Robert Akromond Name : Marrion E. Marxwellson


Index No: 04TU918078 Index No: 04TU918053

supervisor: ING. MICHAEL XENYA


Date : 14/11/2021

EXCELLENCE IN TECHNOLOGY EDUCATION 1


Presentation Outline
❖Background of Study
❖Problem Statement
❖Research Objectives/ Research Questions
❖Significance of Study
❖Literature Review
❖Methodology
❖Test, Results, discussion and Analysis
❖Conclusion And Recommendations

EXCELLENCE IN TECHNOLOGY EDUCATION 2


Background Of The Study
In recent years, the demand for powerful and advanced security monitoring
technologies has increase dramatically.

Computer Vision is a sector of Artificial Intelligence that:


❖ Uses Machine Learning and Deep Learning to allow computers to see,
recognize and analyze things in the same way human eyes do.
❖ Is rapidly gaining popularity for automated quality inspection and
automation.

Trend in computer vision

EXCELLENCE IN TECHNOLOGY EDUCATION 3


Problem Statement
Existing surveillance systems need human vision to analyzed recorded videos
or stored images before an activity is known to take actions.

❖ Human vision can better interpret normal, night, or dark images


better than the computer vision systems in terms of identifying
intruders. Therefore, there is efficiency gap in monitoring system.

❖ It is imperative that surveillance or monitoring systems perform live


detection and make informed decisions in all possible weather and
lighting conditions, running efficient algorithms with the capability to
alert appropriate users.

EXCELLENCE IN TECHNOLOGY EDUCATION 4


Research Objectives
General objective
❖ The main aim of this study is to design and construct a security system with
multiple imaging systems that contain the following features: image
processing, motion detection, and SMS to notify owner. They will be centered
on a Raspberry Pi 4B+ with visible camera, thermal camera and night vision
camera.

Specific objective(s)
❖ Develop a program to process multi-image types in different scenarios
using ImageNet VGG19 Transfer Learning to detect intruders with higher
accuracy and send SMS to a specified mail hub.

EXCELLENCE IN TECHNOLOGY EDUCATION 5


Research Questions

❖ What are the current limitation of similar existing systems?

❖ What techniques are efficient in deploying to resolve the


limitation?

❖ How can the accuracy and effectiveness of the proposed system


be evaluated?

EXCELLENCE IN TECHNOLOGY EDUCATION 6


Significance of Study
In a practical scenario
❖ 10 - 15% extra reliable than current methods in terms video or image
analysis and decision making with no human supervision.
From the perspective of research
❖ This project would contribute to the body of research by serving as a
reference for further studies into the use of advanced Transfer Learning
methods in surveillance monitoring.

From the standpoint of business i.e., companies and enterprises will


make a point to benefit from the intelligent and reliable surveillance
❖ Security Services
❖ Homes
❖ Secured boundaries

EXCELLENCE IN TECHNOLOGY EDUCATION 7


Literature Review
Rev Project Title Author(s) Works Disadvantage(s)
iew
No.
1 Smart border Neha Bhadwal, This project focuses on a The design is not cloud
Surveillance Prateek border surveillance with based and only effective
system Using Agrawal, Vishu intruder detection feature on a small scale with no
Wireless Sensor Madaan (2019) that will run 24/7, automatic detection alert.
Network and enhance the security of
Computer Vision. the state or border area,
utilizing a network of
sensors including passive
infrared detectors and
night vision cameras.
2 CNN-Based Person Jisoo Park, This project offers an It was quite challenging
Detection Using Jingdao Chen, option that provides a to implement since certain
Infrared Images Yong K. Cho, method that comes with a features like object size,
for Night-Time Dae Y. Kang 2 high level of accuracy appearance and
Intrusion Warning and Byung J. with the use of infrared background scenery
Systems. Son (2019) CCTVs in detecting within a dataset cannot
humans at night. be easily generalized for
another use.
EXCELLENCE IN TECHNOLOGY EDUCATION 8
Literature Review Cont’d.
Review Project Title Author(s) Works Disadvantage(s)
No.

3 Thermal Object M. Kristo, M. The goal main of this The design is not
Detection in Ivasic-Kos, reviewed project was cloud based and only
Difficult and M. Pobar, to create or propose effective on a small
Weather
2020 and execute a human scale with no
Conditions
Using YOLO. detection module to automatic detection
detect people or alert.
humans within a
thermal video or
image using
Convolutional Neural
Networks.

EXCELLENCE IN TECHNOLOGY EDUCATION 9


Methodology
Data Collection Methods and Tools
❖ Qualitative data collection method
❖ primary data through simulation

MATLAB 2021a simulation tool

Fig 2 Matlab workspace

EXCELLENCE IN TECHNOLOGY EDUCATION 10


Methodology cont’d.
Transfer Learning
❖ Transfer learning is often considerably faster and easier than training a
network from start with randomly initialized weights for fine-tuning.

Fig 3 Transfer Learning of Pretrained Network for Classification

EXCELLENCE IN TECHNOLOGY EDUCATION 11


Methodology cont’d.
Transfer Network Algorithm with Dataset
Using the Pretrained Deep Neural Network which is trained on Vgg-19 to train and
fine-tune a custom network to suit our dataset and purpose.

Fig 3 Training Dataset for Classification

EXCELLENCE IN TECHNOLOGY EDUCATION 12


Methodology cont’d.
Transfer Network Algorithm with Dataset

Fig 3 Training Process

EXCELLENCE IN TECHNOLOGY EDUCATION 13


Methodology cont’d.
.

Fig 4 operation flow chart.

EXCELLENCE IN TECHNOLOGY EDUCATION 14


System Design And Modelling
Research Design Approach
❖ The assembling of the entire system was initiated by connecting various
components together to achieve the objectives of the study

Fig 5 System Architecture

EXCELLENCE IN TECHNOLOGY EDUCATION 15


System Design
Proposed system

Fig 6 Complete system (opened)

EXCELLENCE IN TECHNOLOGY EDUCATION 16


System Design
Assembled system

Fig 7 Complete system

EXCELLENCE IN TECHNOLOGY EDUCATION 17


Test And Analysis
First Test 1
The detection at day is detected by an LDR located at the top the system. The system uses the
LDR to switch the camera to day mode, and then detection is made.

The image shows the result


of a successful intruder
detection and the fact that
the system cannot
recognize the person as a
user with the indicated box
on the face of the detected
person recognized
Fig 8 Intruder Detection unknown, an intruder.

EXCELLENCE IN TECHNOLOGY EDUCATION 18


Test And Analysis
First Test 2
The detection at day is detected by an LDR located at the top the system. The system uses the
LDR to switch the camera to day mode, and then detection is made.

The figure shows the result


of a successful day, night
and detection and
recognition with the
indicated box on the face of
the detected image
recognized as Akromond, a
user.

Fig 9 Normal, night vision and thermal camera detection

EXCELLENCE IN TECHNOLOGY EDUCATION 19


Conclusion and Limitations
❖ An Intruder Detection System with Alert Using Transfer Learning and Multiple
Imaging Systems was modeled. Intruders and Users were using VGG-19 Trained
Network Algorithm and sends SMS and Email to the registered owner of the system.
The Trained Network Algorithm had 98.4% training accuracy and was exported
from MATLAB to a python framework through ONNX to be implemented to the pi.

Limitations

❖ The main limitation encountered in this work was that the thermal camera used was
of low quality because of the cost of a better thermal camera and limited access on
Twilio account which restricted the usage of MMS. The project experienced
occupational freezing and crashes.

EXCELLENCE IN TECHNOLOGY EDUCATION 20


Recommendations
❖ The following are recommended for future research in the
subject of intruder detection systems employing transfer
learning and multi-imaging systems:

a. It is recommended that a higher microprocessor be used in


future implementation.
b. It is also recommended that higher quality thermal camera
be used.
c. A full subscription of Azure and Twilio is needed to access
effective functionalities.

EXCELLENCE IN TECHNOLOGY EDUCATION 21


References
❖ H. Patel and K. P. Upla, “Night vision surveillance: Object detection using
thermal and visible images,” 2020 Int. Conf. Emerg. Technol. INCET 2020, pp.
8–13, 2020, doi: 10.1109/INCET49848.2020.9154066.

❖ J. Liu, S. Zhang, S. Wang, and D. N. Metaxas, “Multispectral deep neural


networks for pedestrian detection,” Br. Mach. Vis. Conf. 2016, BMVC 2016, vol.
2016-Septe, pp. 73.1-73.13, 2016, doi: 10.5244/c.30.73.

❖ J. Park, J. Chen, Y. K. Cho, D. Y. Kang, and B. J. Son, “CNN-based person


detection using infrared images for night-time intrusion warning systems,”
Sensors (Switzerland), vol. 20, no. 1, pp. 1–15, 2020, doi: 10.3390/s20010034.

EXCELLENCE IN TECHNOLOGY EDUCATION 22


Thank You

EXCELLENCE IN TECHNOLOGY EDUCATION 23

You might also like