Available online at www.sciencedirect.

com
Available online at www.sciencedirect.com
Available online at www.sciencedirect.com
Available online at www.sciencedirect.com

ScienceDirect
Procedia Computer Science 00 (2024) 000–000
Procedia Computer Science 00 (2024) 000–000 www.elsevier.com/locate/procedia
Procedia
Procedia Computer
Computer Science
Science 00 (2024)
235 (2024) 000–000
3369–3378 www.elsevier.com/locate/procedia
www.elsevier.com/locate/procedia
International Conference on Machine Learning and Data Engineering (ICMLDE 2023)
International Conference on Machine Learning and Data Engineering (ICMLDE 2023)
International Conference
International Conference on
on Machine
Machine Learning
Learning and
and Data
Data Engineering
Engineering (ICMLDE
(ICMLDE 2023)
2023)
Edge AI Enabled IoT Framework for Secure Smart Home
Edge AI
Edge AI Enabled
Enabled IoTIoT Framework
Framework for for Secure Smart Home
Infrastructure Secure Smart Home
Infrastructure
a Infrastructure
Pankaj Thakur , Shubham Goela,∗, Emjee Puthooranb
Pankaj Thakuraa , Shubham Goela,∗ a,∗
, Emjee Puthooranbb
Pankaj Thakur ,
a Computer Science
a
Shubham
and Engineering,Goel , Emjee
JUIT, Waknaghat, Puthooran
and Solan, India
Computer
b Electronics andScience and Engineering,
Communication JUIT,JUIT,
Engineering, Waknaghat, and Solan,
Waknaghat, IndiaIndia
and Solan,
b a Computer Science and Engineering, JUIT, Waknaghat, and Solan, India
Electronics and Communication Engineering, JUIT, Waknaghat, and Solan, India
b Electronics and Communication Engineering, JUIT, Waknaghat, and Solan, India

Abstract
Abstract
Abstract
The synergy of Edge Computing, Artificial Intelligence, and Internet of Things in smart city camera-based surveillance offers
The synergy
notable of Edge
advantages Computing,
in resource Artificial
allocation. Intelligence,
Traditional and Internet
approaches of Things
involving in video
constant smart data
city streaming
camera-based surveillance
to central servers offers
incur
The synergy
notable of Edge
advantages Computing,
in resource Artificial
allocation. Intelligence,
Traditional and Internet
approaches of Things in video
smart data
city streaming
camera-basedcentral
surveillance offers
significant bandwidth and storage costs. Our contribution involvesinvolving
proposing constant
a cost-effective, lightweighttoEdge AI servers
Enabledincur
IoT
notable advantages in resource
significant allocation.Our
Traditional approaches involving constant video data lightweight
streaming to central AI servers incur
Frameworkbandwidth
for Secureand storage
Smart Homecosts. contribution
Infrastructure, utilisinginvolves proposing
a Raspberry a cost-effective,
Pi single-board computer and the Edge Enabled
open-source IoT
software
significant
Framework bandwidth
for Secure and storage
Smart Home costs. Our contribution involves proposing a cost-effective, lightweight Edge AI Enabled IoT
motion for camera surveillance. TheInfrastructure,
motion program utilising
monitorsa Raspberry Pi single-board
video signals from various computer
camerasand andthe open-source
triggers specificsoftware
actions
Framework
motion for Secure
for camera Smart Home
surveillance. TheInfrastructure,
motionefficiently
programutilising
monitorsa Raspberry
video Pi single-board
signals computer and the open-source software
upon detecting movement. The framework notifies the smart homefrom various
owner cameras
via email andand triggers
smartphone specific
messageactions
when
motion
upon for camera surveillance. The motion program monitors video signals from various cameras and triggers specific actions
motiondetecting movement.
is detected. The framework
We integrated efficiently
four advanced motion notifies the smart
detection homenotification
and alert owner viamethodologies,
email and smartphone
conductingmessage when
a thorough
upon
motion detecting movement.
is detected. The framework
We integrated efficiently
four advanced motionnotifies the smart
detection homenotification
and alert owner via methodologies,
email and smartphone
conductingmessage when
a thorough
evaluation that positioned our framework as superior to existing solutions. Our research showcases impressive accuracy rates of
motion is detected.
evaluation We integrated
thatinpositioned ouroutdoorfour advanced
framework motion
as superior detection
to existing and alertOur
solutions. notification
research methodologies,
showcases conducting a thorough
impressive
91% and 85% indoor and scenarios, with minimal average delays of 12.8 seconds for email alerts andaccuracy ratesfor
1.6 seconds of
evaluation
91% and 85%that positioned
inisindoor our framework
and outdoor as superior to existing solutions. Our research showcases impressive accuracy rates for
of
messages which approximately 41.6scenarios,
% less thanwith
theminimal
state of average delays of 12.8 This
the art methodologies. seconds for email
innovative alerts and
integration 1.6only
not seconds
elevates
91% and 85%
messages which in isindoor and outdoor
approximately 41.6scenarios,
% less with
than theminimal
state ofaverage
the delays of 12.8 seconds for email alerts andnot
1.6only
seconds for
surveillance capabilities but also establishes a swift and reliable alertartsystem,
methodologies. Thissignificantly
contributing innovative integration
to the efficiency and elevates
security
messages
surveillancewhich is approximately
capabilities 41.6 % less
but also establishes than the
a swift andstate of the
reliable artsystem,
alert methodologies. Thissignificantly
contributing innovative integration not only
to the efficiency and elevates
security
of smart home environments.
surveillance
of smart home capabilities
environments. but also establishes a swift and reliable alert system, contributing significantly to the efficiency and security
of smart home environments.
© 2024 The Authors. Published by Elsevier B.V.
©
© 2024
This is an
2024 The
TheopenAuthors. Published
accessPublished
Authors. by Elsevier
article under
by B.V.
the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
© 2024
This is The
an openAuthors.
access Published
article by ELSEVIER
under Elsevier
the CC B.V.B.V.
BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Peer-review
This is an open under
accessresponsibility
article underof the scientific
CC BY-NC-ND committee of the
license International Conference on Machine Learning and Data Engi-
(https://creativecommons.org/licenses/by-nc-nd/4.0)
This is an open
Peer-review access article under
under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Peer-review
neering. underresponsibility
responsibilityofofthethescientific
scientificcommittee
committee of the International
of the Conference
International on Machine
Conference Learning
on Machine and Data
Learning andEngi-
Data
Peer-review
neering.
Engineering under responsibility of the scientific committee of the International Conference on Machine Learning and Data Engi-
neering.
Keywords: Edge AI; IoT framework; Smart home; surveillance; motion.
Keywords: Edge AI; IoT framework; Smart home; surveillance; motion.
Keywords: Edge AI; IoT framework; Smart home; surveillance; motion.

1. Introduction
1. Introduction
1. Introduction
The convergence of Edge computing (EC), Artificial Intelligence (AI), and the Internet of Things (IoT) represents
The convergence
a transformative of Edge
paradigm computing
shift (EC),
in the realm Artificial
of Artificial
technology Intelligence (AI), and the
and data processing. Internet
This ofisThings
mergerof (IoT)
redefining represents
how we col-
The convergence
alect,
transformative of Edge
paradigm computing
shift in the (EC),
realm of technology Intelligence
and data (AI), and the
processing. Internet
This merger isThings (IoT)
redefining represents
how we col-
alect, process, and
transformative utilize
paradigm information in real-time,
shift in theinrealm with profound
of technology implications
and data for
processing.for industries,
This merger isbusinesses, and
redefiningand everyday
howeveryday
we col-
process,
life. The and
fusion utilize information real-time, with profound implications industries, businesses,
lect, process,
life. The andof
fusion
EC, AI,
ofutilize
EC, AI,
and the IoT
information
and the IoTin is usheringwith
real-time,
is
in a new era ofimplications
ushering in a profound
smart city camera-based
for industries,surveillance
new era of smart city camera-based businesses, that
surveillance
promises
and everyday
that promises
life. The fusion of EC, AI, and the IoT is ushering in a new era of smart city camera-based surveillance that promises
∗Corresponding author. Email.: [email protected], [email protected]
∗ Corresponding
∗
E-mail address:author. Email.: [email protected],
[email protected], [email protected]
[email protected], [email protected]
Corresponding
E-mail address:author. Email.: [email protected],
[email protected], [email protected]
[email protected], [email protected]
E-mail address:
1877-0509 [email protected],
© 2024 The Authors. [email protected],
by Elsevier B.V. [email protected]
1877-0509
This
1877-0509 ©
© 2024
is an open Thearticle
access
2024 The Authors. Published
under
Authors. by ELSEVIER
Elsevier B.V.
the CC BY-NC-ND
Published by license
B.V.(http://creativecommons.org/licenses/by-nc-nd/4.0/)
1877-0509
This isisan ©under
2024responsibility
Thearticle
Authors. Published by Elsevier B.V.
This anopen
Peer-review openaccess
access under
article of
under CC
the BY-NC-ND
the scientific license
committee
CC BY-NC-ND (http://creativecommons.org/licenses/by-nc-nd/4.0/)
oflicense
the International Conference on Machine Learning and Data Engineering.
(https://creativecommons.org/licenses/by-nc-nd/4.0)
This is an open
Peer-review access
under article under
responsibility the scientific
of the CC BY-NC-ND license
committee of (http://creativecommons.org/licenses/by-nc-nd/4.0/)
the International Conference on Machine Learning and Data Engineering.
Peer-review under responsibility of the scientific committee of the International Conference on Machine Learning and
Peer-review under responsibility of the scientific committee of the International Conference on Machine Learning and Data Engineering.
Data Engineering
10.1016/j.procs.2024.04.317
3370 Pankaj Thakur et al. / Procedia Computer Science 235 (2024) 3369–3378
2 Author name / Procedia Computer Science 00 (2024) 000–000

to revolutionize urban safety and security. This convergence equips surveillance cameras with the ability to process
data locally, analyze it in real-time, and make intelligent decisions, greatly enhancing the effectiveness and efficiency
of city-wide surveillance systems.
At the heart of this transformation is EC, which shifts the data processing from centralized servers to the vicinity
of the surveillance cameras themselves. Instead of transmitting vast amounts of video footage to a remote data center
for analysis, EC enables these cameras to process the data they capture locally, near the source. This localization min-
imizes latency and reduces the load on network bandwidth, allowing for rapid decision-making and more responsive
surveillance.
The incorporation of AI into smart city camera-based surveillance brings a wealth of intelligence to the system.
AI algorithms, particularly those driven by machine learning, enable the cameras to recognize and interpret what they
see in real-time. This includes the ability to identify and track objects, detect unusual behavior or anomalies, and even
analyze facial expressions or license plates. This level of intelligence significantly enhances the surveillance system’s
ability to identify potential threats and respond proactively. The relationship between AI and EC can indeed be mutu-
ally beneficial [12] (as shown in Fig. 1), as they can complement each other to enhance overall system performance
and meet specific application requirements. Consider a scenario where a smart city camera equipped with EC and AI
capabilities detects an abandoned bag [4] in a crowded public space. Instead of relying on human operators to notice
this anomaly, the camera can autonomously recognize it as a potential security concern. It can trigger immediate alerts
to law enforcement or security personnel, drastically reducing response times and enhancing public safety.

Fig. 1: Mutually beneficial relationship between AI and EC [12].

The IoT aspect of this convergence adds a layer of connectivity and data-sharing capabilities to smart city surveil-
lance. Surveillance cameras can be interconnected with other IoT devices, such as environmental sensors, traffic man-
agement systems, and emergency response mechanisms. For instance, a camera detecting a sudden traffic accident
[15] can promptly alert traffic management systems to reroute vehicles and reduce congestion, while simultaneously
notifying emergency services for immediate response. Moreover, IoT-enabled surveillance can adapt to dynamic sit-
uations. For example, during a natural disaster or civil unrest [13], cameras can collaborate with other IoT sensors
to assess the severity of the situation. They can then dynamically adjust their surveillance focus to provide relevant
information to first responders, helping them make informed decisions in real-time. Another vital aspect of this conver-
gence is privacy and data security. With localized processing, surveillance footage is less vulnerable to cyberattacks
during data transit to centralized servers. Additionally, AI algorithms can be employed to perform real-time video
anonymization [20], ensuring the privacy of citizens and complying with data protection regulations. Only relevant
information, such as potential security threats, needs to be transmitted to central authorities, enhancing data security
and privacy. Furthermore, this approach offers scalability. As the smart city surveillance network grows, additional
cameras and IoT devices can be seamlessly integrated into the existing infrastructure. This scalability enables cities
to continually adapt and expand their surveillance capabilities to meet evolving security needs.
As the drive to bolster the capabilities of EC and AI in IoT continues to gain momentum, it becomes apparent that
within the context of smart cities and the Cloud-Edge-Things continuum, application level motion detection remains
a relatively underexplored area, currently lacking comprehensive coverage through open-source programs like motion
[24], for motion detection, and the MQTT (Message Queue Telemetry Transport) based Mosquitto broker [9], for
 Pankaj Thakur et al. / Procedia Computer Science 235 (2024) 3369–3378 3371
Author name / Procedia Computer Science 00 (2024) 000–000 3

sending lightweight and secure alert notifications to the user, to the best of our current knowledge. In this work, our
contribution to selected research field in terms of utilizing a surveillance camera connected to a Raspberry Pi Single
Board Computer (SBC) installed with the motion [24], an open source program for motion detection, to develop an
Edge AI Enabled IoT Framework for Secure Smart Home Infrastructure. Using this framework the smart home owner
will be notified through an email and a message on his/her smartphone when a motion is detected in front of the
camera. Moreover, in our contribution to AI aspect of current research work, we have also taken care of unnecessary
alerts due to moments by birds or insects. The mails will be sent by the sendmail, an open source program running
on the Raspberry Pi while the messages will be sent by the Mosquitto broker program running on an Amazon Web
Services (AWS) based Ubuntu Server instance.
The subsequent sections of this paper are structured as follows: Section 2 provides an in depth exploration of related
works. In Section 3, we delve into the details of our proposed edge AI enabled framework designed to enhance the
security of smart home infrastructure. Section 4 comprehensively addresses the results and outlines the experimental
setup employed. Lastly, Section 5 encapsulates the conclusion and delineates avenues for future work.

2. Related Work

Numerous solutions for motion detection and subsequent user notifications have been proposed, with many of them
relying heavily on the use of Passive Infrared (PIR) sensor as a primary method for motion detection. Table 1 offers a
comprehensive comparative analysis of these proposed solutions.
In the realm of home security systems outlined in various studies, innovative configurations leverage Raspberry Pi,
PIR sensors, and cameras, often incorporating OpenCV and advanced detection mechanisms. For instance, Kumar et
al. (2017) in [16] integrates OpenCV with Haar cascade for motion detection, triggering email and message notifica-
tions upon sensing motion through a PIR sensor, while authors in [18] employs a PIR sensor connected to Raspberry
Pi, capturing images and videos with the NoIR Pi camera, and relaying notifications, including media files, to the user
via the Pushbullet API. Average delay in receiving the alert notification is also cited in the work. Facial recognition
takes center stage in [27], where a camera module, a PIR sensor and Raspberry Pi collaborate for this purpose, while
[25] introduces an IoT-based security system using PIR sensors, Pi Camera, and Raspberry Pi for motion detection,
with accessible images on mobile phones. [1] presents a smart surveillance system with Raspberry Pi and PIR sen-
sors at the doorstep, capturing intruder images and verifying access rights. Innovative approaches abound, such as
[32], where switch sensors beneath floor tiles detect motion, triggering the Raspberry Pi to activate the camera. [21]
disguises a security system as a smart mirror, capturing intruder photos and notifying the homeowner via the Pytho-
nAnywhere cloud and the SMSLane service. [14] employs NodeMCU-32S with an ultrasonic distance sensor and the
Blynk application for an advanced security system, while In [23], the authors discusses an intelligent home security
system using an Orange Pi, camera, GSM module, and Android application. The system captures and analyses in-
truder images, comparing them to a database of known individuals. When a match is found, the homeowner receives
a notification with the intruder’s name and image. If there’s no match, relevant information from the database is sent.
The introduced systems exhibit diversity, with [26] dynamically adjusting sensor activity based on user presence, [22]
utilizing face authentication for a door lock system, and [11] employing PIR sensors, NodeMCU, and Raspberry Pi
for a surveillance monitoring system.
Robotic solutions emerge in [30], where a robot follows a path and signals the homeowner upon detecting intruders.
In [2], the authors introduce an affordable IoT-based home security system. When motion is detected by the PIR sensor,
it triggers a buzzer and the ESP32-CAM camera module mounted atop a servo motor. The GSM Module connected to
an Arduino UNO, sends alert messages to the homeowner. [8] uses Raspberry Pi, camera, and PIR sensors to capture
and notify via Telegram. Integrating smart features, [29] links Google Assistant to home security, adjusting sensor
states based on homeowner presence. Alert messages are sent to the owner via the Blynk Application over Wi-Fi. In
cases of no internet connectivity, messages are routed through the GSM Module. The work demonstrated in [31] uses
the PIR sensor to detect any motion and a camera to capture the intruder’s image. The system uses ESP8266 along
with machine learning algorithms to find intruders. The user is notified by email and mobile application. The security
system discussed in [19] integrates a PIR sensor and a laser beam with Light Dependent Resistance (LDR). When the
PIR sensor detects motion or an intruder crosses the laser beam, an Arduino Nano activates the alarm, WiFi camera,
and communicates to the NodeMCU, which then forwards the message on the mobile application. In [17] the face
3372 Pankaj Thakur et al. / Procedia Computer Science 235 (2024) 3369–3378
4 Author name / Procedia Computer Science 00 (2024) 000–000

Table 1: Comparison of related works

Citation Control Unit Motion Detection Notification

[16] Raspberry Pi PIR sensor Email and message
[18] Raspberry Pi PIR sensor Pushbullet based notification
[27] Raspberry Pi PIR Sensor Email, LED
[25] Raspberry Pi PIR sensor Raspicam Remote App
[1] Raspberry Pi PIR sensor Android App
[32] Raspberry Pi switch sensor Notification
[21] Raspberry Pi Camera SMSLane service based notifi-
cation
[14] NodeMCU32S Ultrasonic sensor Blynk App
[23] Orange Pi Camera Android App, Call
[26] ESP-32, ESP32- PIR sensor, Reed Sensor Buzzer Notification
CAM, Arduino
Mega
[22] Raspberry Pi Camera -
[11] Raspberry Pi and PIR sensor -
NodeMCU
[30] Raspberry Pi and PIR sensor and Ultrasonic Buzzer, SMS
NodeMCU sensor
[2] Arduino UNO and PIR sensor Android App
Node MCU
[8] Raspberry Pi PIR sensor Telegram
[29] Node MCU PIR sensor Email, Blynk App
[31] ESP8266 PIR sensor Email, Mobile Application
[19] Arduino Nano, Ar- PIR sensor and LDR Android Application
duno Mega and Node
MCU
[17] Arduino Mega, Ultrasonic sensor Blynk, Bolt Mobile Applica-
ESP32-CAM, tions
NodeMCU
[10] Raspberry Pi Camera Notification
[5] Raspberry Pi Camera Buzzer, SMS
[3] Raspberry Pi PIR Sensor Telegram
[7] NodeMCU and Ar- Ultrasonic Sensor Notification
duino UNO
[28] Arduino UNO and PIR Sensor Notification
Raspberry Pi
[6] Raspberry Pi Ultrasonic Sensor Cayenne App
Proposed Raspberry Pi Camera Email, Message (on mobile
Framework App)

recognition system utilizing the Blynk and Bolt Mobile Application have been demonstrated. It makes use of multiple
ultrasonic sensors to detect intrusion. Authors in [10] use a camera in a house connected to a Raspberry Pi enabling the
capture of movement and objects. This data is subsequently sent to a Fog layer for verification, determining whether
the detected object poses a threat. If a harmful object is identified, the user receives an alert notification. It uses a web
based interface in the cloud for viewing the captured images.
 Pankaj Thakur et al. / Procedia Computer Science 235 (2024) 3369–3378 3373
Author name / Procedia Computer Science 00 (2024) 000–000 5

The system demonstrated in [5] uses OpenCV, Raspberry Pi and a camera to identify and classify objects using
image processing tools, extracting information from real time video feeds. When human movement is detected, the
system promptly sends alert notifications to the user via SMS. In [3], when the PIR sensor connected to a Raspberry
Pi detects motion the Pi Camera records a 30-second video and sends to the Botfather for notifying the user through
Telegram. In [7], the authors introduces a smart security system capable of detecting visitors within 5 feet of the
doorstep using an ultrasonic sensor. Additionally, a thermal camera is employed to capture thermal images, allowing
the system to determine the temperature of the visitor through image processing. The collected data is then uploaded
to the cloud using NodeMCU. When a visitor is detected in the red zone, the homeowner receives an alert notification.
Users can access visitor information, including the visit date, time, and the visitor’s body temperature, providing
enhanced security and awareness. In [28], the authors introduced a robot equipped with an ultrasonic sensor and a
PIR sensor which patrols the house when the user is away. Upon detecting motion via the PIR sensor, the Arduino
board activates the Raspberry Pi, initiating the web camera to capture images within the house. These images are then
transmitted to the user through a cloud based system, providing remote monitoring and enhancing home security. In
[6], the paper presents a smart security system comprising an ultrasonic sensor, Raspberry Pi, buzzer, relay module,
and CCTV camera. When the ultrasonic sensor detects motion, the Raspberry Pi activates a buzzer and CCTV camera.
It uses the Cayenne application to send notifications to the user. Users can access live images from the CCTV camera
via the Mi Home application.
The PIR sensors, utilised in many of the above cited works, have inherent constraints associated for motion detec-
tion. This encompassing issues like limited detection range, susceptibility to environmental conditions, and challenges
in adapting to non-thermal motion. These factors may collectively lead to less precise readings. Furthermore, the chal-
lenge of dealing with undesired notifications triggered by birds in outdoor camera scenarios and by spiders, bugs, and
insects in indoor camera setups has not received sufficient attention in the current body of literature.
To address these limitations effectively, we have opted for a camera-based motion detection system. This alternative
not only overcomes the aforementioned shortcomings but also offers a wide array of benefits, including an expansive
range of motion detection capabilities, the ability to capture intricate visual data, provide extensive coverage, and
enable seamless remote monitoring. This strategic choice ensures the provision of a robust solution, particularly in
scenarios where the acquisition of critical visual information is indispensable. Also, within the dynamic landscape
of smart cities and the evolving Cloud-Edge-Things continuum, the domain of application-level motion detection
stands out as a relatively underexplored territory. Presently, this crucial aspect lacks thorough coverage from available
open-source programs such as motion [24] and the Mosquitto broker [9], at least as per our current understanding.

3. Proposed Framework

The proposed framework, illustrated in Fig. 2, is designed with a primary focus on real-time motion detection. The
Table 2 provides a concise overview of the hardware and software configuration embedded within the proposed frame-
work. The central aim is to swiftly alert homeowners to any identified motion, guaranteeing heightened awareness and
facilitating an immediate response.
In its hardware setup the framework leverages the Raspberry Pi board as the central processing unit for motion
detection. The Raspberry Pi is attached to a camera, which captures video footage of the its surroundings. In the
software part we used the motion [24], a Linux-based motion detection program on the Raspberry Pi to process the
video stream in real-time. This software is configured to detect any significant changes in the camera feed, such as
motion or object presence. The AI module is embedded in the motion program so that unnecessary and irritating alert
notifications are avoided.
To notify the owner of the smart home we have used two notification mechanisms: email and message. Upon
detecting motion in front of the camera, the motion program discerns alterations in pixels. Depending on the predefined
threshold value for this pixel change, it start video recording and then triggers a Linux bash script. This script utilizes
the Linux sendmail program to dispatch an email to the homeowner’s registered email address and concurrently
publishes an alert message on a designated topic within the MQTT Mosquitto broker hosted on an Amazon Elastic
Compute Cloud (Amazon EC2) instance. The system records video footage for a duration of 60 seconds, storing
it in the memory card connected to the Raspberry Pi. Additionally, when motion is detected, the system captures
a single optimal still image of the event. The email provides crucial information, including the timestamp of the
3374 Pankaj Thakur et al. / Procedia Computer Science 235 (2024) 3369–3378
6 Author name / Procedia Computer Science 00 (2024) 000–000

Fig. 2: Proposed Framework.

Table 2: Environmental Setup

Sr. Item Brief configuration

No.
1 AWS EC2 in- t2.micro instance having 1GB RAM, Ubuntu Server 22.04 LTS,
stance mosquitto MQTT Broker version 2.0.11.
2 IP Camera CP Plus 2MP camera, model CP-UNC-DS21PL3.
3 Single Board Raspberry Pi 3B+ model having 1GB RAM, Ubuntu Server 22.04, mo-
Computer tion 4.5.1, sendmail and 32 GB SanDisk class 10 memory card.

event and a snapshot from the camera feed. This immediate email notification ensures that homeowners are alerted
to potential security concerns, even when they are not actively monitoring their smart home. In addition to email
notifications, our framework includes a dedicated Android mobile application installed on the homeowner’s mobile
device. When motion is detected, an alert notification is also sent to the owner’s smartphone using the Mosquitto
[9], a MQTT protocol based message broker. This notification includes the timestamp of the motion detection for
real-time monitoring. The mobile application provides homeowners with a convenient and user-friendly interface for
managing security alerts. Fig. 3 provides a comprehensive flowchart within which the entire operation of the proposed
framework is elucidated.
Traditional surveillance cameras have enjoyed decades of widespread use, recording footage for subsequent review
by human operators. Nevertheless, advancements in motion [24] have enabled the acceptance of existing movie files
as an input, facilitating the generation of notifications in response to detected motion events.

4. Results and Experimental Setup

Our proposed motion detection framework underwent rigorous testing involving 100 instances (60 with motion,
40 without) everyday for a period of 30 days, between 06:00 AM to 11:00 PM for both indoor and outdoor video
sequences and generated the confusion matrix each day for each of these 100 events. We meticulously curated these
events to encompass diverse scenarios, including varying light conditions, the presence of living beings, varying
distance from camera and different weather conditions, thereby rigorously evaluating the system’s resilience and
performance.
Throughout these tests, we consistently maintained the frame rate at a steady 30 frames per second and the image
dimensions for all video frames were standardised to dimensions of 640 pixels in width and 480 pixels in height. In
 Pankaj Thakur et al. / Procedia Computer Science 235 (2024) 3369–3378 3375
Author name / Procedia Computer Science 00 (2024) 000–000 7

Fig. 3: Flowchart for (a) motion program running on the Raspberry Pi (b) mosquitto MQTT Broker running on Amazon EC2 (c) mosquitto
subscriber running on the user’s smartphone.

Fig. 4: Confusion Matrix for single day (a) Indoor video sequences (b) Outdoor video sequences.

Fig. 4, we present the confusion matrix for these motion detection events for single day. True Positives (TP) represent
the instances accurately identified as motion, while True Negatives (TN) denote instances correctly recognized as non-
motion. On the other hand, False Positives (FP) signify instances mistakenly categorised as motion when they are, in
fact, non-motion. Conversely, False Negatives (FN) indicate instances wrongly classified as non-motion when they
are, indeed, in motion. For our proposed framework, False Negatives (FN) resulting from Type-II errors are of greater
concern than False Positives (FP) associated with Type-I errors. This is because FN instances signify the framework’s
failure to detect actual motion.
The works cited in Section 2, ”Related Work,” predominantly overlook the issue of delays in alert notifications
associated with receiving emails, messages, and other relevant details. In order to address this research gap compre-
hensively, we executed implementations guided by the methodologies delineated in [16], [25], and [14]. Summary
of the state of the art methodologies for motion detection are presented in Table 3. Subsequently, we compared the
resulting delays and other performance metrics with those found in [18] and our proposed framework. The nuanced
findings are meticulously presented in Fig. 5. Contrasting our proposed framework with state-of-the-art methodolo-
gies involves evaluating accuracy, precision, and recall across distinct setups. Specifically, (a), (c), and (e) present
metrics for indoor scenarios, while (b), (d), and (f) depict corresponding values for outdoor environments. In addition,
Figures (g) and (h) provide insights into the average delay experienced in alert notifications for 25 motion events,
encompassing both email and message notifications. These evaluations are conducted over a comprehensive 30-day
period for both indoor and outdoor video sequences.
For the indoor video sequence, our findings reveal an impressive accuracy rate of 91%, complemented by a pre-
cision of 89% and a robust recall of 96%. Likewise, in the context of the outdoor video sequence, our framework
attains an accuracy level of 85%, accompanied by a precision and recall standing at 82% and 93%, respectively. These
3376 Pankaj Thakur et al. / Procedia Computer Science 235 (2024) 3369–3378
8 Author name / Procedia Computer Science 00 (2024) 000–000

(a) (b)

(c) (d)

(e) (f)

(g) (h)

Fig. 5: Comparison of state of art methodologies M-I: [18], M-II: [16], M-III: [25] and M-IV: [14] with our proposed framework: (a), (c), and (e)
present metrics for indoor scenarios, while (b), (d), and (f) depict corresponding values for outdoor environments. Subfigures (g) and (h) provide
insights into the average delay experienced in alert notifications for 25 motion events per day over a period of 30 days.
 Pankaj Thakur et al. / Procedia Computer Science 235 (2024) 3369–3378 3377
Author name / Procedia Computer Science 00 (2024) 000–000 9

Table 3: Summary of State of the art methodologies for motion detection.

Sr. Methodology Description

No.
1 Methodology-I It was presented by Majumder et al. [18], in which when the motion is detected
(PIR-RPi-NoIR) by a PIR sensor connected to the Raspberry Pi, the NoIR Pi camera capture
images and videos. Alert notification are sent through Pushbullet API.
2 Methodology-II It was presented by Kumar et al. [16], in which upon sensing motion through
(PIR-RPi-Camera) PIR sensor, the camera promptly captures and image and trigers email and
message notifications to alert the user.
3 Methodology-III It was presented by Rao et al. [25], in which when the motion is detected by a
(PIR-RPi-Pi) PIR sensor connected to the Raspberry Pi, the Pi camera capture image of the
motion. It uses the Raspicam Remote App
4 Methodology-IV It was given by Jotawar et al. [14], in which upon sensing motion through ultra-
(Ultrasonic- sonic distance sensor, the camera captures image and notifies the user through
NodeMCU-Camera) the Blynk server.

findings underscore the effectiveness and reliability of our framework in accurately detecting motion across a range
of indoor and outdoor environments.

5. Conclusion and Future work

In this study, we have successfully demonstrated the advantages of integrating Edge Computing, Artificial Intelli-
gence, and Internet of Things in the context of smart city camera-based surveillance. By leveraging EC and AI-driven
local analysis, we have significantly reduced the need for continuous data transmission. This has led to a substantial
decrease in network congestion, as well as associated costs related to bandwidth and storage resources. Our framework
for secure smart home infrastructure is cost-effective, making it accessible to a wide range of users. The utilisation
of open-source software further contributes to affordability. The smart homeowners are promptly notified via email
and smartphone messages with an average delay of 12.8 and 1.6 seconds respectively, whenever motion is detected
in front of the surveillance camera. The suggested framework possesses the capability to mitigate superfluous and
bothersome alert notifications triggered by the abrupt presence of birds and insects.This real-time alerting system
enhances security and situational awareness.
While our study has provided valuable insights and contributions to the integration of EC, AI, and IoT in smart
city surveillance, there are several promising avenues for future research and development. Further research can focus
on enhancing the AI algorithms used for real-time analysis to improve object recognition, behaviour analysis, and
anomaly detection which may lead to more accurate and efficient surveillance systems. Investigating methods to
scale our framework for larger deployments, including smart city-wide surveillance networks may also be taken for
future study. This involves addressing challenges related to data synchronisation, resource management, and security.
Also developing energy-efficient algorithms and hardware solutions for EC devices is crucial, especially for battery-
powered IoT devices. Privacy-preserving AI models and encryption techniques to ensure the security of video data and
personal information, exploring ways to enhance the compatibility and interoperability of EC devices and IoT sensors
for seamless integration into existing smart city infrastructures, investigating the use of machine learning models to
predict and prevent security incidents rather than merely responding to detected motion to improve proactive security
measures are also the need of hour.

References

[1] Advirkar, S., Bhatkar, P.V., Katke, N.S., Ghosal, D., 2020. Smart surveillance system. International Journal of Research in Engineering,
Science and Management (IJERSM) 3, 70–72.
3378 Pankaj Thakur et al. / Procedia Computer Science 235 (2024) 3369–3378
10 Author name / Procedia Computer Science 00 (2024) 000–000

[2] Ahmed, T., Nuruddin, A.T.B., Latif, A.B., Arnob, S.S., Rahman, R., 2020. A real-time controlled closed loop iot based home surveillance
system for android using firebase, in: 2020 6th International Conference on Control, Automation and Robotics (ICCAR), IEEE. pp. 601–606.
[3] Azhar, A.H., Othman, M.F.I., Bahaman, N., Mas’ud, M.Z., Sa’aya, Z., et al., 2021. Implementation of home security motion detector using
raspberry pi and pir sensor. Journal of Advanced Computing Technology and Application (JACTA) 3, 41–50.
[4] Borkar, A., Nagmode, M., Pimplaskar, D., 2013. Real time abandoned bag detection using opencv. International Journal of Scientific &
Engineering Research 4, 660.
[5] Chetan, B., Bharath, P., Akarsh, S., Vernerkar, M., Swamy, B., 2021. Smart surveillance system using tensor flow. International Journal of
Innovative Research in Electrical, Electronics, Instrumentation and Control Engineering (IJIREEICE) 9, 96–99.
[6] Chong, P.L., Ganesan, S., Than, Y.Y., Ravi, P., 2022. Designing an autonomous triggering control system via motion detection for iot based
smart home surveillance cctv camera. Malaysian Journal of Science and Advanced Technology , 80–88.
[7] Das, S., Dhar, A., Pal, B., Biswas, P., Gayen, P.K., Majumder, S., 2021. Design and development of iot based smart security system in covid19
situation, in: Journal of Physics: Conference Series, IOP Publishing. p. 012048.
[8] Desnanjaya, I., Arsana, I.N.A., 2021. Home security monitoring system with iot-based raspberry pi. Indones. J. Electr. Eng. Comput. Sci 22,
1295.
[9] Foundation, E., . Eclipse mosquitto an open source mqtt broker. https://mosquitto.org/. (Accessed Sep. 08 2023).
[10] Gladence, M., Revathy, S., Jeyanthi, P., et al., 2021. Security management in smart home environment .
[11] Harun Al Rasyid, M.U., Astika Saputra, F., Kurniawan, A., 2020. Surveillance monitoring system based on internet of things, in: 2020
International Electronics Symposium (IES), pp. 588–593. doi:10.1109/IES50839.2020.9231634.
[12] Hua, H., Li, Y., Wang, T., Dong, N., Li, W., Cao, J., 2023. Edge computing with artificial intelligence: A machine learning perspective. ACM
Computing Surveys 55, 1–35.
[13] Huszár, V.D., Adhikarla, V.K., Négyesi, I., Krasznay, C., 2023. Toward fast and accurate violence detection for automated video surveillance
applications. IEEE Access 11, 18772–18793.
[14] Jotawar, D., Karoli, K., Biradar, M., Pyruth, N., 2020. Iot based smart security and home automation. Int. Res. J. Eng. Technol.(IRJET) 7.
[15] Keote, M., Bhosale, S., Sargar, A., Kumbhare, S., Kukade, M., Kontamwar, G., 2023. Ai camera for tracking road accidents, in: 2023 7th Inter-
national Conference on Intelligent Computing and Control Systems (ICICCS), pp. 565–568. doi:10.1109/ICICCS56967.2023.10142418.
[16] Kumar, K.K., Natraj, H., Jacob, T.P., 2017. Motion activated security camera using raspberry pi, in: 2017 International Conference on Com-
munication and Signal Processing (ICCSP), IEEE. pp. 1598–1601.
[17] Lulla, G., Kumar, A., Pole, G., Deshmukh, G., 2021. Iot based smart security and surveillance system, in: 2021 international conference on
emerging smart computing and informatics (ESCI), IEEE. pp. 385–390.
[18] Majumder, A.J., Izaguirre, J.A., 2020. A smart iot security system for smart-home using motion detection and facial recognition, in: 2020
IEEE 44th Annual Computers, Software, and Applications Conference (COMPSAC), IEEE. pp. 1065–1071.
[19] Murad, M., Bayat, O., Marhoon, H.M., 2021. Design and implementation of a smart home system with two levels of security based on iot
technology .
[20] Myneni, S., Agrawal, G., Deng, Y., Chowdhary, A., Vadnere, N., Huang, D., 2022. Scvs: On ai and edge clouds enabled privacy-preserved
smart-city video surveillance services. ACM Trans. Internet Things 3. URL: https://doi.org/10.1145/3542953, doi:10.1145/
3542953.
[21] Nadafa, R.A., Hatturea, S., Bonala, V.M., Naikb, S.P., 2020. Home security against human intrusion using raspberry pi. Procedia Computer
Science 167, 1811–1820.
[22] Paul, J., Bhowmick, R.S., Das, B., Sikdar, B.K., 2020. A smart home security system in low computing iot environment, in: 2020 IEEE 17th
India council international conference (INDICON), IEEE. pp. 1–7.
[23] Phaltanwala, F., Banawadikar, S., Burse, S., Bhutada, K., 2020. Smart home security system. International Research Journal of Engineering
and Technology (IRJET) 7, 6444–6447.
[24] Projects, M., . Motion. https://motion-project.github.io/index.html. (Accessed Sep. 08 2023).
[25] Rao, B.N., Sudheer, R., 2020. Surveillance camera using iot and raspberry pi, in: 2020 Second International Conference on Inventive Research
in Computing Applications (ICIRCA), IEEE. pp. 1172–1176.
[26] Razali, M.N.M., Abdul-Kadir, N.A., Kasim, J., 2020. Smart home security with dual modes, in: 2020 IEEE Student Conference on Research
and Development (SCOReD), IEEE. pp. 453–458.
[27] Sanjay, A., Vijarania, M., Jaglan, V., 2020. Security surveillance and home automation system using iot. EAI Endorsed Transactions on Smart
Cities 5.
[28] Sendhilkumar, N., Malarvizhi, C., Anand, M., Periyarselvam, K., 2021. Internet of things based indoor smart surveillance and monitoring
system using arduino and raspberry pi, in: Journal of Physics: Conference Series, IOP Publishing. p. 062083.
[29] Shirisha, E., et al., 2021. Iot based home security and automation using google assistant. Turkish Journal of Computer and Mathematics
Education (TURCOMAT) 12, 117–122.
[30] Sunitha, M., Vinay, P.D., Lokesh, V., Kumar, B.D., 2020. Ip based surveillance robot using iot, in: 2020 fourth international conference on
I-SMAC (IoT in Social, Mobile, Analytics and Cloud)(I-SMAC), IEEE. pp. 337–342.
[31] Taiwo, O., Ezugwu, A.E., 2021. Internet of things-based intelligent smart home control system. Security and Communication Networks 2021,
1–17.
[32] Venugopal, N., Kumar, C.E.P., Prashanth, V.G., Manoharan, E., Kesavamurthy, K., 2020. Iot based security system using raspberry pi. Inter-
national Journal of Engineering Research and Technology (IJERT) 9.