PiBase: An IoT-based Security System Using

Google Firebase and Raspberry Pi

Venkat Margapuri Niketa Penumajji Mitchell Neilsen
Department of Computer Science Department of Computer Science Department of Computer Science
Kansas State University Kansas State University Kansas State University
Manhattan, KS Manhattan, KS Manhattan, KS
[email protected] [email protected] [email protected]

Abstract— Smart environments are environments where digital learning. The application can also be used in other smart
devices are connected to each other over the Internet and operate in environments where authorized access is required. The
sync. Security is of paramount importance in such environments. ecosystem of the application consists of a Raspberry Pi, a
This paper addresses aspects of authorized access and intruder Passive Infrared (PIR) motion sensor, a Raspberry Pi camera
detection for “smart” environments. Proposed is PiBase, an Internet module and an Android mobile phone. The software for the
of Things (IoT)-based app that aids in detecting intruders and system is developed in Java, Python and NodeJS. The user
providing security. The hardware for the application consists of a information is stored in Google Firebase, a cloud environment
Raspberry Pi, a PIR motion sensor to detect motion from infrared
provided by Google. The use of Google’s software solution
radiation in the environment, an Android mobile phone and a
ensures that user information is protected at all times and the
camera. The software for the application is written in Java, Python
and NodeJS. The PIR sensor and Pi camera module connected to the
information is robust to attacks compared with other home-
Raspberry Pi aid in detecting human intrusion. Machine learning grown solutions.
algorithms, namely Haar-feature based cascade classifiers and In the remainder of the article, Section II describes related
Linear Binary Pattern Histograms (LBPH), are used for face work, Section III describes the hardware and software
detection and face recognition, respectively. The app lets the user components, Section IV, the architecture and implementation
create a list of non-intruders and anyone that’s not on the list is and Section V sheds light on future work and concludes the
identified as an intruder. The app alerts the user only in the event of
article.
an intrusion by using the Google Firebase’s Cloud Messaging
service to trigger a notification to the app. The user may choose to II. RELATED WORK
add the detected intruder to the list of non-intruders through the app
to avoid further detections as intruder. Face detection by the Haar Chowdhury, Nooman and Sarker [5] proposed an access
Cascade algorithm yields a recall of 94.6%. Thus, the system is both control system using Raspberry Pi for home security. The
highly effective and relatively low cost. application is setup to send an email and tweet to the user
anytime a face is detected at the door. The user is required to
Keywords— Android, Google Firebase, Haar features, Linear re-tweet with the door lock or open command to trigger the
Binary Pattern Histogram, machine learning, PIR sensor magnetic lock which in turn opens or keeps the door locked.

I. INTRODUCTION Pavithra and Balakrishnan [14] developed an

implementation of IoT to monitor and control home
One of the major concerns for homes and businesses in the appliances using a Raspberry Pi, a PIR sensor and a
modern world is the threat of property intrusion and burglary. smartphone via the Internet. Automated control of lights, fans
There are 2.5 million burglaries in the US annually according and fire detection in the event of a fire are demonstrated in the
to the US Department of Justice yearly data and over half of work.
them are home invasions [19]. An increased number of people
in the work-force limits the amount of time they spend at home Sajjad et al. [18] devised a face recognition framework
leaving homes vulnerable. Besides, the rise of e-commerce using Raspberry Pi for enhanced law-enforcement services.
during the COVID-19 pandemic has led to a substantial rise in The proposed approach used Viola Jones approach for face
products shipped to homes and establishments. It is reported detection. Besides, the system used Bag of Words technique
that 35.5 million people in the US alone have been victims of for extraction of oriented FAST and rotated BRIEF points
porch pirating with the average stolen package costing $156 from the detected face, followed by support vector machine
[9]. The need to safeguard homes and establishments is for identification of suspects.
paramount and security solutions are the need of the hour. The Prathaban, Thean and Sazali [15] developed a vision-
article presents one such solution using the notion of Internet based home security system using Raspberry Pi and OpenCV.
of Things (IoT). IoT may be perceived of as the network of The proposed solution leveraged Haar-Cascade algorithm
connected devices that communicate over the Internet while coupled with background subtraction and Histogram of
following a certain set of protocols to achieve a common goal. Oriented Gradients (HOG). The solution was tested against a
In other words, individual “smart” entities connected over the system that only contained PIR sensors. It was observed that
Internet form a “smart” network and operate in harmony. the OpenCV solution had an accuracy of 100% as opposed to
Security cameras act as a formidable means to prevent and the PIR sensor system’s 76%.
provide evidence of property intrusion after the fact. Dash and Nayak [13] implemented a smart surveillance
Generally, city dwellers place security camera systems that monitoring system using Raspberry Pi and PIR sensor for
cost anywhere between a few hundred to a few thousand mobile devices. Notifications to the user’s smartphone are sent
dollars to safeguard their homes. This article proposes PiBase, over the internet upon the detection of motion to alert the user.
a low-cost, machine-to-machine (M2M) IoT application that
aids in securing homes and establishments using machine

XXX-X-XXXX-XXXX-X/XX/$XX.00 ©20XX IEEE

 Nadaf, Hatture, Bonal and Naik [12] proposed a system down resistors are used to do so. Only pins
designed as a smart mirror that provides both home security GPIO2 and GPIO3 have fixed pull-up resistors
and other information such as news, weather information and but the others may be configured in software.
calendar. The application leveraged the Amazon cloud to store
the images captured and SMLLane servers to send SMS B. PIR Motion Sensor
notifications to the user. PIR motion sensors work on the principle of
Ansari, Sedky, Sharma and Tyagi [3] developed a security Pyroelectricity i.e. the property of materials where they
system using Raspberry Pi, camera and PIR motion sensor to generate a certain amount of voltage when subject to
monitor and trigger alarms when motion is detected. The temperature changes. The PIR sensor consists of two slots,
captured photos and videos are sent directly to a remote cloud each one made of a material that is sensitive to infrared (IR)
server. In the event the cloud server is unavailable, they are radiation and lens. Both slots have the ability to detect motion
stored on the Raspberry Pi temporarily until the cloud server within the sensitivity range of the sensors. In the idle state
becomes available. where the sensors are exposed to ambient conditions, either
slot detects the same amount of IR radiation. However, when
Abaya, Basa, Sy, Abad and Dadios [1] designed a security
an object approaches the sensor, it intercepts one half of the
system using Raspberry Pi and OpenCV. The developed
PIR sensor that causes a positive differential change between
system was put to test inside a warehouse facility. In the
absence of a dedicated motion detection sensor, the system the two halves. As the object leaves the sensing area, the
used background subtraction algorithm for motion detection. opposite happens and the sensor generates a negative
Detection for humans was performed using Haar features and differential change. The positive and negative differential
triggered e-mail notification, an LED indicator and sound changes trigger an output pulse of 3V. The lens around the
alarm. In the event the detected motion was not a human, the sensors is a Fresnel lens. The use of Fresnel lens reduces the
system checked for smoke by using the property, ‘the amount of material required compared to a conventional lens
direction of hot smoke is always upward in the absence of since it divides the lens into a set of concentric annular
wind’. sections.
Gupta, Patil, Kadam and Dumbre [7] explored the C. Pi Camera Module
feasibility of implementing Raspberry Pi face detection and The Pi Camera module is the camera provided for the
recognition using conventional face detection and recognition Raspberry Pi by the Raspberry Pi Foundation. The camera
techniques such as Haar detection and PCA. The objective of used as part of the work is Camera Module v2, an 8-
the work was to propose a solution that could replace the use
megapixel camera first released in 2016.
of passwords and RF I-Cards for access to high security
The software developed for the application is written in Java,
systems and buildings besides lowering the investment in
them. Python and NodeJS. Each of the software components is as
described further.
Patil and Shukla [16] implemented a classroom attendance
system based on face recognition in class using Raspberry Pi. A. Android Application
The Viola Jones algorithm was used for face detection and The Android application acts as the UI for PiBase and is
face recognition using Principal Component Analysis (PCA) implemented in Java. The app provides the ability for the user
and Linear Discriminant Analysis (LDA). to register, login, upload a list of intruders and view any
intrusions that might occur. Google Firebase is used to
III. HARDWARE AND SOFTWARE COMPONENTS authenticate and store any information entered by the user on
The hardware components include a Raspberry Pi 3.0, PIR the app such as the list of intruders. The notifications of
motion sensor, Pi Camera module and Android smartphone. intrusion which include a picture and timestamp are are
The components are briefly described below. received on the app. Initially, the notifications are displayed
on Android’s notification tray but the user may click on the
A. Raspberry Pi 3.0
notification to view it in the app.
The Raspberry Pi 3.0 is a small-sized computer that has
the ability to plug into a computer monitor or any other B. Google Firebase
display using HDMI. It runs the Raspbian OS and is handy to One of the cloud services provided by Google is Firebase.
run applications in programming languages such as Python. The environment provides a multitude of services such as
The Raspberry Pi is provided with a 40-pin General Purpose user authentication and security, cloud to client messaging,
Input/ Output (GPIO) that helps to interact with external database, emulator suite, hosting for micro-services, storage
components such as the PIR motion sensor. The voltages and for user-generated content such as images and videos and
the input/ output capabilities of the pins are as follows: plenty of other features. The ones pertinent to the current
• Voltages: Two 5V pins and two 3V pins are work are user authentication, database and cloud messaging.
present on the Raspberry Pi along with a number User Authentication: Firebase provides a complete
of ground pins which are not configurable. The backend solution for authentication to sign-in with
remainder of the pins are general purpose 3V passwords, federated identity providers, email links and text
pins meaning inputs are 3V tolerant and outputs messages. The Admin SDK shipped as part of Firebase helps
are 3V. to integrate authentication services into custom apps built in
• Output: A GPIO pin designated as output pin can Android (or other platforms). The user may read and write
be set to high (3V) or low (0V). data to the real-time database, send cloud messages, generate
• Input: A GPIO pin designated as an input pin can and verify Firebase auth tokens using the Admin SDK.
be set to high (3V) or low (0V). Pull-up and pull-
 Database: Firebase provides users with a NoSQL real- Cascade, the features are applied in stages where the stages
time database where data is synced across all clients in real- in the beginning comprise of simpler features whereas the
time, and remains available at all times. The data in the features in the later stages comprise of complex features. In
database is stored in JSON. The clients are able to use case a feature fails on a particular window in an image, the
references to the JSON data to write data and subscribe to algorithm concludes that no facial features are present in the
data changes made by other clients attached to the database. image and the remaining features are not run on the image.
The database also provides users with the ability to write The algorithm progresses in the aforementioned manner
security rules that regulate access to the data eventually detecting a face.
Storage: Firebase Storage provides users with the ability Linear Binary Pattern Histogram (LBPH): LBPH is a
to store and serve user-generated content, such as images or technique that builds on top of the technique of Linear Binary
videos. The files are stored in a Google Cloud Storage bucket Patterns (LBP). LBP labels the pixels of an image by
that makes them accessible through Firebase and Google thresholding the neighborhood of each pixel and considers
Cloud. Cloud Storage SDK is available to access and the result as a binary number. The idea behind LBPH is that
manipulate the stored data. The SDK integrates with Firebase LBP when combined with histograms represents facial
Authentication as well making it easier to setup user- images as a simple data vector. The first step in LBPH is the
based/group-based security. creation of a master list with images that the algorithm needs
Firebase Cloud Messaging (FCM): FCM is a cross- to recognize. A unique ID is needed to be assigned to all the
platform cloud-messaging service that lets users send images that belong to a certain individual. A window of a
messages from Firebase to clients. FCM is able to send certain dimension (say 3 x 3) is extracted from the grayscale
messages to an individual client app or place messages on a representation of the original image. The window may be
‘topic’. A topic may be understood as a shared represented as a matrix with each cell containing the intensity
communication channel wherein the messages placed in a of a pixel. The central value in the matrix is considered the
topic are available to all the clients that are subscribed to the threshold. For each of the neighbors of the threshold, a value
topic, essentially enabling a publish-subscribe mechanism. of 1 is assigned if the value is greater than or equal to the
The messages sent by FCM are classified into two categories: threshold, 0 otherwise, resulting in a matrix containing only
Notification and Data. The primary difference between the binary values. The binary values in the matrix are
two categories is that Notification messages are displayed concatenated line-by-line in clockwise direction and
automatically on the user’s console, be it a smartphone or converted to a decimal value. The decimal value is assigned
tablet whereas Data messages are required to be processed by to the location of the threshold in the matrix. This process
the client application. Besides, Notification messages have a when repeated by sliding the window across the image results
predefined set of user-visible keys and optional data payload in a new image that better represents the characteristics of the
of custom key-value pairs whereas Data messages only original image. Grid parameters i.e. the predefined cells in the
contain custom key-value pairs with no reserved key names. horizontal and vertical direction (usually 8 x 8) are used to
Both message types have a maximum payload limit of 4000 divide the image into multiple grids and a histogram is
bytes. The Admin SDK may be used to integrate and interact generated for each grid in the image. Each of the histograms
with FCM. is then concatenated resulting in a large histogram that
represents the characteristics of the original image. In order
C. Face Detection and Face Recognition
to recognize a face, the LBPH algorithm is run and the
Face detection and recognition are performed using resulting histogram is compared against the histograms of
OpenCV, an open-source computer vision library. Face each of the images in the master list to identify the best match.
detection is performed using the Haar Cascades algorithm
and recognition is performed using Linear Binary Pattern IV. ARCHITECTURE AND IMPLEMENTATION
Histograms (LBPH) algorithm. Both algorithms are popular
A. Architecture
in the area of computer vision and are briefly described.
Haar Cascades: Haar Cascades, otherwise known as the Figure 1 shows the modularized architecture of PiBase.
Viola-Jones algorithm is an object detection algorithm to The system comprises of two communication channels
established over the internet – one between the Android
identify faces in images and videos. The algorithm leverages
smartphone and Google Firebase, another between Google
the proposed Haar Features that aid in detecting edges or lines
Firebase and Raspberry Pi.
on the images, or to identify areas on the image with varying
intensities of pixels. The exhaustive set of features proposed Dataflow from Android App to Raspberry Pi: In the channel
by the Viola Jones algorithm contains 180,000 features. between the Android smartphone and Google Firebase, the
However, not every feature holds relevance to the task in Android smartphone leverages the authentication system
question. Adaboost is a feature selection technique that helps provided by Google Firebase to register and authenticate users
to constrict the set of features. Adaboost applies each of the on PiBase. Besides, the user is allowed to upload images of
180,000 features to the images in the dataset separately and people that PiBase may perceive of as non-intruders. The
creates weak learners. Weak learners are defined as images are stored in the Storage of Google Firebase and a
classifiers that perform better than random guessing but still corresponding record is inserted in the real-time database. On
the second channel between Google Firebase and Raspberry
exhibit performance that is below-par in terms of classifying
Pi, the Raspberry Pi pulls down the images uploaded by the
positive and negative images. The resulting feature set Android app. The images received by the Raspberry Pi are
consists of fewer features eliminating the ones that are used to train the face recognition algorithm.
irrelevant. Finally, a technique known as Attentional Cascade
is applied on each of the images to detect faces. In Attentional
Dataflow from Raspberry Pi to Android App: Upon the non-intruders are added to the database. In order to view past
detection of an intruder, the Raspberry Pi uploads the image intrusions, the user may pick a start date and end date. The
to the real-time database of Google Firebase. Upon intrusions within the date boundary are displayed as a list on
recognizing a write to the real-time database, Google Firebase the app. On the flip side, Google Firebase sends notifications
triggers a message to each of the users subscribed to receive to the app anytime an intrusion is detected by Raspberry Pi.
notifications of the intrusion. The notifications are visible on Android’s notification tray as
B. Implementation shown in Figure 3(c). The user may click on the notification
to view in the app.
As evident from the architecture, the system consists of
b. Google Firebase
three major components – Android smartphone, Google
The backend services for the application are provided by
Firebase and Raspberry Pi. The primary use case of the
Google Firebase, a cloud environment. The platform Admin
application is to detect intruders and notify users. The
SDK provided by the platform is used to perform user
implementation pertinent to each of the components is as
authentication through an Android app. The
described below.
signInWithEmailAndPassword() method is used to pass the
a. Android Application
email and password credentials of the user from Android. The
The Android application is implemented in Java on
method has callback listeners which indicate the success or
Android Studio integrated development environment (IDE).
failure of the login. Another key feature of Firebase’s that is
The flow of the application on the smartphone is as shown in
leveraged is Firebase Cloud Messaging. As mentioned in
Figure 2. The app is connected to Google Firebase using the
section III, Google Firebase is able to send notifications to
Google Firebase Console since the app relies on Firebase’s
users. Since all of the apps connected to a Firebase project are
services for authentication and database. Firstly, the user is
required to be notified right away upon the detection of an
required to register with PiBase where the user’s name, email
intrusion which involves a write to the real-time database,
and password of their choice are collected as shown in Figure
Firebase Cloud Functions are used. Cloud Functions is a
3(a). The credentials are sent to Google Firebase which
serverless framework that executes a predefined set of
creates a new account linked to the credentials. The
instructions or code in response to an event. A cloud function
information is stored as part of the Firebase project and in the
that is triggered in response to a write/ update on the database
event multiple apps are connected to a single Firebase project,
is setup on Google Firebase. Figure 4 shows a snippet of the
the credentials may be used to authenticate the user on any
trigger that sends a data message to the Android app upon
app of the project. Upon registration, the user may use the
receiving an intrusion image from the Raspberry Pi on a topic
registration credentials to login to the application. Upon a
named ‘rpi_security’.
successful login, the user is presented with options to upload
c. Raspberry Pi
name and images of non-intruders or view past intrusions.
The Raspberry Pi 3.0 is equipped with the Pi Camera
Multiple images of each of the non-intruders are desired to
module and PIR motion sensor. Figure 5 shows the circuit
aid in detection. The images and names of each of the non-
diagram of the PIR motion sensor and Pi Camera module
intruders are added to the upload screen as shown in Figure
connected to the Raspberry Pi. The camera is inserted into the
3(b). Clicking on the ‘Choose Image’ button lets the user
camera module port of the Raspberry Pi. The camera’s blue
select images from the smartphone’s filesystem. The user
color strip needs to face the ethernet port. For the PIR motion
may view the next or previous images by clicking on the
sensor, the voltage pin may be inserted into any one of the
‘Next Image’ or ‘Previous Image’ button respectively. The
DC voltage (5V) pins of the Raspberry Pi, the ground (GND)
uploaded images are sent to storage on Google Firebase and
pin may be connected to any one of the GND pins on the
stored in a folder with the name provided on the upload
Raspberry Pi and Digital OUT pin to one of the GPIO input
screen. Besides, a record containing the folder name of the
pins on the Raspberry Pi. The PIR motion sensor and camera
inserted images and timestamp is inserted in the real-time
act in tandem to detect intruders. Since the Raspberry Pi is
database which the Raspberry Pi later uses to check if new

Figure 1: PiBase Architecture Figure 2: Android Smartphone Dataflow

 Figure 3: (a) User Registration Screen; (b) Non-Intruder Image Upload Screen; (c) Intrusion Notification from Google Firebase
resource constrained, running both PIR motion sensor and time frame. Besides, the automation script initiates
camera at all times hogs the resources for the Raspberry Pi to training of the LBPH algorithm on the images. Training the
perform other tasks. In order to prevent it, only the PIR LBPH algorithm right after the Android app uploads a non-
motion sensor is in operation at all times. The camera is only intruder is desired. However, training the LBPH algorithm
activated upon a detection of motion by the PIR sensor. The consumes resources on the resource-constrained Raspberry
dataflow of PiBase on the Raspberry Pi is as shown in Figure Pi and training the algorithm multiple times a day might
6. Since the goal is to detect and recognize an intruder, of the hinder the performance of the device. One of the features of
essence is an image where the face is clearly visible. LBPH is that a confidence score is output along with the
However, the person in the vicinity of the camera moves and prediction. Unlike in the conventional sense, a low
does not always face the camera head-on. So, face detection confidence score is desired. In case LBPH does not match the
and recognition are not an easy task. In order to best achieve image against any of the images in the master list, the
an image with a clear face, the camera is setup to capture ten detected image is uploaded from the Raspberry Pi to the real-
images with an interval of two seconds between each image time database in Google Firebase.
in hopes that at least one of the images shows the face clearly.
C. Experiment
Practically, it helps to have an enticing object below the
Raspberry Pi setup as bait to attract the intruder towards the The Raspberry Pi is installed inside a home environment
camera. Each of the captured images is run through the Haar and tested on a real-world dataset containing a master list of
Cascades face detection algorithm. In the event a face is 23 people with fifteen males and eight females. Six images of
detected, the face is put through the LBPH face recognition each people’s faces are captured with the head in different
algorithm to identify the face. The LBPH algorithm is orientations. When put to test wherein each of the 23 people
required to be trained on the non-intruder image dataset is made to take a stroll inside the room four to five times, the
before it can be put to the task of facial recognition. In order Haar Cascades face detection algorithm detects 106 faces
to do so, the Raspberry Pi pulls down the images from Google over a total of 112 trials. The metrics of Precision and Recall
Firebase once a day. Since the Android app inserts a record are used to evaluate the face detector. Precision refers to the
containing the folder name and timestamp in the real-time number of positive correctly identified by the classifier and is
database, an automation script written in Python and given by 𝑡𝑟𝑢𝑒 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠/ (𝑡𝑟𝑢𝑒 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠 +
triggered daily on the Raspberry Pi retrieves non-intruder 𝑓𝑎𝑙𝑠𝑒 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠) whereas Recall refers to the number of
images based on the records inserted during the past 24-hour positives identified of all the positives in the dataset and is

Figure 5: Raspberry Pi Connected to Pi Camera and PIR

Figure 4: Firebase Trigger on Real-time Database Motion Sensor [17]
 confidence scores for males with a peculiar beard pattern are
lower (better) than males without beards.
V. FUTURE WORK AND CONCLUSION
While face detection and recognition are performed
decently well, there is room for improvement. In future, the
idea is experiment with pre-trained and home-grown
architectures of deep neural networks meant for resource
constrained devices such as the Raspberry Pi. Additional
functionality such as an alarm system that may be triggered
Lower Bound Upper Bound Recall (%) from the Android app is in consideration.
0 45 86% Overall, PiBase is a low-cost security system
10 75 91% implemented on Android and Raspberry Pi with Google
10 100 94% Firebase acting as the backend. The system demonstrates the
10 150 100% feasibility of IoT and mobile applications to security systems.
While the proposed system is applied in terms of home
security, the idea of face detection and recognition may be
applied in any field where the need for authorized access
persists.
REFERENCES
[1] Abaya, W. F., Basa, J., Sy, M., Abad, A. C., & Dadios, E. P. (2014,
November). Low cost smart security camera with night vision
Figure 6: Raspberry Pi 3.0 Dataflow capability using Raspberry Pi and OpenCV. In 2014 International
conference on humanoid, nanotechnology, information technology,
given by 𝑡𝑟𝑢𝑒 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠/(𝑡𝑟𝑢𝑒 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠 + communication and control, environment and management
𝑓𝑎𝑙𝑠𝑒 𝑛𝑒𝑔𝑎𝑡𝑖𝑣𝑒𝑠). The face detector exhibits a Precision of (HNICEM) (pp. 1-6). IEEE.
1.0 since no false positives are detected. The recall is 94.64%. [2] Abuzneid, M. A., & Mahmood, A. (2018). Enhanced human face
The detected images are run through the LBPH face recognition using LBPH descriptor, multi-KNN, and back-propagation
neural network. IEEE access, 6, 20641-20651.
recognizer. The face recognizer does not misclassify an
[3] Ansari, A. N., Sedky, M., Sharma, N., & Tyagi, A. (2015, January). An
individual i.e. no false positives. Hence, the Precision is 1.0. Internet of things approach for motion detection using Raspberry Pi.
The Recall at different thresholds of confidence scores is as In Proceedings of 2015 International Conference on Intelligent
shown in Table 1. Computing and Internet of Things (pp. 131-134). IEEE.
Table 1: Face Recognition Recall of LBPH [4] Chatterjee, N., Chakraborty, S., Decosta, A., & Nath, A. (2018). Real-
time communication application based on android using Google
Figure 7 shows the workflow of PiBase’s image capture, face firebase. Int. J. Adv. Res. Comput. Sci. Manag. Stud, 6(4).
detection and face recognition on an individual in the vicinity [5] Chowdhury, M. N., Nooman, M. S., & Sarker, S. (2013). Access
Control of Door and Home Security by Raspberry Pi Through
of the camera. Inferring from the results in Table 1, the Internet. Int. J. Sci. Eng. Res, 4(1), 550-558.
percentage of recognition improves as the bounds are [6] Cuimei, L., Zhiliang, Q., Nan, J., & Jianhua, W. (2017, October).
loosened. While thresholds are user defined and may be Human face detection algorithm via Haar cascade classifier combined
loosened at will, it is imperative that the algorithm learns to with three additional classifiers. In 2017 13th IEEE International
Conference on Electronic Measurement & Instruments (ICEMI) (pp.
recognize the faces with a high level of certainty. From the 483-487). IEEE.
experiments, the observation is that the face recognizer works [7] Gupta, I., Patil, V., Kadam, C., & Dumbre, S. (2016, December). Face
best when the entities in question have distinct features. detection and recognition using Raspberry Pi. In 2016 IEEE
Facial hair seems to have an impact since the recognition international WIE conference on electrical and computer engineering
(WIECON-ECE) (pp. 83-86). IEEE.
[8] Khawas, C., & Shah, P. (2018). Application of firebase in android app
development-a study. International Journal of Computer
Applications, 179(46), 49-53.
[9] Laycock, R & Choi, C. (2020). Porch Pirates Statistics.
https://www.finder.com/porch-pirates-statistics
[10] Moroney, L. (2017). Firebase cloud messaging. In The Definitive
Guide to Firebase (pp. 163-188). Apress, Berkeley, CA.
[11] Margapuri, V. (2020). Smart Motion Detection System using
Raspberry Pi. arXiv preprint arXiv:2006.06442.
[12] Nadaf, R. A., Hatture, S., Challigidad, P. S., & Bonal, V. M. (2019,
June). Smart mirror using raspberry pi for human monitoring and home
security. In International Conference on Advanced Informatics for
Computing Research (pp. 96-106). Springer, Singapore.
[13] Nayak, M., & Dash, P. (2014). Smart surveillance monitoring system
using Raspberry Pi and PIR sensor.
[14] Pavithra, D., & Balakrishnan, R. (2015, April). IoT based monitoring
and control system for home automation. In 2015 global conference on
communication technologies (GCCT) (pp. 169-173). IEEE.
[15] Prathaban, T., Thean, W., & Sazali, M. I. S. M. (2019, November). A
Figure 7: PiBase Image Capture and Face Recognition vision-based home security system using OpenCV on Raspberry Pi 3.
 In AIP Conference Proceedings (Vol. 2173, No. 1, p. 020013). AIP
Publishing LLC.
[16] Patil, A., & Shukla, M. (2014). Implementation of classroom
attendance system based on face recognition in class. International
Journal of Advances in Engineering & Technology, 7(3), 974.
[17] Raspberry Pi Projects. (2017). Raspberry Pi Surveillance with Linux
and Azure. https://projects-raspberry.com/raspberry-pi-surveillance-
camera-linux-azure/
[18] Sajjad, M., Nasir, M., Muhammad, K., Khan, S., Jan, Z., Sangaiah, A.
K., Sangaiah & Baik, S. W. (2020). Raspberry Pi assisted face
recognition framework for enhanced law-enforcement services in smart
cities. Future Generation Computer Systems, 108, 995-1007.
[19] Stasha, S (2021). 27 Alarming Burglary Statistics for 2021.
https://policyadvice.net/insurance/insights/burglary-statistics/