Cost Effective AI-based Screening System for COVID-19

Shiva Madhav Adusumilli1, Priya Valentina Albert1, Rohit Boddeda1, Samhita L

Kolluri 1, Satwik Chowdary2, Jyothi Putti3
1
CSE, VNR Vignana Jyothi Institute of Engineering &Technology, Hyderabad, India
2
MECH, VNR Vignana Jyothi Institute of Engineering &Technology, Hyderabad, India
3
Asst. Professor at Vignana Jyothi Institute of Engineering &Technology, Hyderabad, India

shivaadusumilli5@gmail.com, priya.valentina21@gmail.com,
rohitboddeda@gmail.com, samhita2310@gmail.com,
satwikchowdary17@gmail.com, jyothi_p@vnrvjiet.in

Abstract. For over a year, the world is still battling a global pandemic caused
by the novel corona virus. Leveraging the technical potential has become
crucial in ensuring safety with the emergence of new strains. In this paper, we
propose a low-cost AI-based screening system that can be deployed at various
locations. The objective of our solution is to automate the task of face mask
detection, check for social distancing and body temperature scanning. The
dataset consisted of 2314 images combining those without a mask and those
with an artificial mask attached to them. These three functionalities are carried
out by combining AI and IoT technologies. Specifically, we employed the
SingleShot-Multibox detector (SSD) and ResNet-10 architecture as our first
model for face detection. MobileNetV2 architecture is used as our second
model for binary classification (with or without mask). Various components are
combined to achieve the real-time screening process task and subsequently
granting access through a controlled door. Our proposed model outperformed
other existing solutions by achieving an accuracy of 99% and an F1 score of
0.99.

Keywords: COVID-19, Deep Learning, Convolution, Screening system, Raspberry Pi,

Serial Peripheral Interface (SPI), inter-integrated circuits (IIC), Universal
asynchronous receiver-transmitter (UART).

1 Introduction
The covid-19 pandemic, caused by the SARS-COV-2 virus, has been ravaging across
the country, causing disruption and unprecedented chaos to all spheres of life. It has
resulted in millions of lives lost, a plummeting economy, the overwhelming burden on
the healthcare workers, underwhelming health infrastructure to support, and thereby
wrecking the livelihood of many citizens. With the emergence of new variants, the
corona virus still poses a significant threat to our health today, despite the
development of vaccines. Slow vaccination programs will take a long time to cater to
the huge Indian population before reducing the infection rates substantially and
achieving herd immunity. Moreover, the efficacy of these vaccines against new
looming variants has been drawing skepticism from experts lately. Although vaccines
2

have proven to provide a safety net against severe infection and eventual death, these
mutating variants are spreading faster than ever leading to a staggering number of
deaths. In the interim of getting the whole nation vaccinated, it becomes all the more
essential to follow the prescribed protocols. Frequent hand washing, wearing face
masks, and maintaining social distance in crowded places can mitigate the rate of
transmission and scope for evolving variants.
Eighteen months into the pandemic, many technological solutions have been
deployed in public places to ensure the public's safety. Security systems have been
adopting various state-of-the-art health surveillance tools. Face mask detection
cameras and temperature sensing have become a necessity in all public places. The
manual method of screening using an infrared thermometer involving a human
standby has proven ineffective and time-consuming. Hence people are highly
susceptible to contracting the virus in densely populated areas. Public places like
airports and commercial spaces often employ expensive automated systems. Although
AI has been integrated to automate screening at the entrance of public places, we have
not achieved the goal of scalability and economic viability. These systems are costly
and resource-intensive for small-scale deployments like offices, residential
complexes, and private spaces. Hence, the scope to improvise is endless.
Our paper intends to employ IoT devices with deep learning and computer vision
algorithms to achieve a real-time covid screening solution. Our objective is to provide
a cost-effective solution for automating face mask detection and temperature sensing
by leveraging IoT. Components like Raspberry Pi, IR Temperature Module, Servo
Motors, Pi camera, Solenoid Lock, Motion Detector, and Relay will be used to
achieve the task. For image classification, a pre-trained MobileNetV2 architecture is
used. The metrics used to test the model are accuracy and loss. The working of the
proposed system will be further discussed in detail.

2 Related Work
The work mentioned in [1] developed a CNN model for mask detection using the
images extracted from the footage of CCTV cameras. The images of the violators
were sent via the smart network system to the concerned authorities. It achieved an
accuracy of 98.7% accuracy and an AUC of 0.985. The study in [2] built a machine
learning model for face mask detection using MobileNetV2 architecture. They used
the Real-Word Masked Face dataset (RMFD) and images collected from 25 cities in
proportion to their population to correlate the vigilance index of Covid-19 with the
percentage of people wearing masks. This study showed an accuracy of 96.85%. The
SSDMNV2 model was proposed in [3] that implemented a Single Shot Multibox
Detector for face detection, OpenCV deep learning techniques for easier real-time
embedded applications, and MobileNetV2 architecture image classification. An
accuracy of 92.64% and an F1 score of 0.93 were achieved. MobileNetV2
architecture was concluded to be optimal as it outperformed other alternatives in
terms of accuracy and f1 score.
A comprehensive survey of how IoT was used to provide holistic healthcare and
specifically combat the covid-19 pandemic was clearly articulated in [4]. Another
rule-based approach to assist in remote covid diagnosis and treatment was adopting
 3

wearable sensors to provide physiological data about the patient and monitored using
the smart application introduced in [5].
A recent innovation [6] has integrated IoT technology, thermal and optical cameras
in a smart helmet to screen and obtain real-time data about people with high
temperatures. Furthermore, a humanoid IoT-based solution that combined various
modules like IR Sensor, Autonomous Navigation, Camera Module, CT for chest,
sensor kit, Camera Module for capturing, and Blood Sample collection was
mentioned in covid-19 diagnoses and identification [7].

3 Proposed Methodology
In most of the works discussed in the literature survey, only one or two technologies
were used to achieve a definite functionality. In our work, we propose a more
comprehensive solution that is not just cost-effective but also offers the benefits of
providing varied features that make up an efficient screening system. Our goal is to
design a system that can automate temperature scanning, face mask detection, and
check for social distancing violations when entering an office or commercial space.
This system can be installed at the entrances and connected to an existing or a new
electronic door. Various components are utilized for the whole screening process. The
operational procedure of the proposed screening system is elucidated below.

As the person approaches the entrance, the system automatically gets triggered
with the help of a motion sensor. It checks if social distancing is being followed or
not, and subsequently, a voice and display message are given to the user to maintain
distancing. If not violated, then the system captures the image of the first person in the
queue and extracts features using deep learning techniques to identify a mask's
presence. The absence of a mask will give out a warning message. The system then
proceeds to scan the temperature of the individual with an IR sensor. If the
temperature falls within the specified range, the door is opened for a few seconds to
grant access. Finally, all logs of captured images are stored in a local database for
future reference. This process is repeated, and the whole processing for each person
takes 1 second, thereby overcoming the limitation of time constraint.

3.1 Proposed AI module

Dataset description
Our dataset has 2314 images consisting of 1220 people wearing a mask and 1094
images without a mask. Due to the non-availability of the dataset containing masks,
we have created the dataset by taking an ordinary image of faces (As shown in Fig1)
and an artificial mask image. This artificial mask is applied to the image using various
computer vision techniques (As shown in Fig 2). The facial features such as mouth,
nose, and chin are considered the dominant facial landmarks in the image. By
applying the bounding box to the ordinary image, we acquire the precise location of
the face. In the next step, with the help of Numpy and OpenCV modules, we extracted
the facial features and located the region of interest (ROI). Once the features are
detected, the artificial mask image with no background is rotated and resized to fit the
4

facial features in the ordinary image appropriately. The dataset is increased using
different data augmentation techniques such as horizontal flip, vertical flip, and
rotation of the images to improve the model's generalization ability. In our dataset, the
reference images used for generating the masked images are not considered for
training to eliminate bias and decrease generalization error. Similarly, different types
of datasets can be created by localizing the facial landmarks.

Fig.1 Image without mask Fig.2 Image with artificial mask

Image preprocessing
Initially, we used ImageDataGenerator to generate images in various orientations to
create a batch of images. These generated batches contain the normalized data, which
speeds up the learning process of the model. The LabelBinarizer preprocessing
technique is used to gather the essential features from the image and generalize the
model very accurately. Our system first collects a frame from the video stream and is
fed into two models after reshaping the image size to 224x224 with three channels
(RBG). In the next step, we convert these image pixels to an array by ensuring the
intensity of each pixel in the input falls between the range [-1, 1], and then these
preprocessed images are appended to the data with their corresponding binary labels
in the form of lists. The labels are encoded by using the One-Hot Encoding method to
achieve the best classification performance.

Architecture description
In this paper, we have used two different models. The first model is responsible for
locating the faces in a frame and subsequently checking for social distancing among
the located faces of individuals. The second model is to classify the image whether
the specific individual is wearing a mask. In the first model, the face location is
detected using various OpenCV techniques. The new version of OpenCV contains a
Deep Neural Network (DNN) module that can support different frameworks such as
Tensor Flow, Caffe, and PyTorch with pre-trained weights, making it easier to load
models from the disk. The DNN module takes the loaded model as an input and
selects the backend process. The loaded model is built using the SingleShot-Multibox
detector (SSD) for face detection and ResNet-10 architecture as the backbone. After
loading the model with the pre-trained weights, the DNN model detects the face from
the frame with a resized width shape of 400.
 5

Fig.3 This illustrates the layers of MoblieNetV2 architecture [8].

The second model is used for the binary classification of images into individuals
with and without masks. We have used MobileNetV2 architecture as our base model
with pre-trained ImageNet weights. The MobileNetV2 comes from the family of
MobileNet [8][9][10]. The MobileNet was a breakthrough in mobile vision
applications such as mobile devices having memory and power constraints.
MobileNet uses depthwise separable convolutions and is built to take lightweight
DNN's by introducing two global hyper parameters. The MobileNet's are applied to
many tasks like object detection, fine-grain classification, face attributes, and
landmark recognition. MobileNetV1 uses depthwise separable convolutions and point
convolutions by scaling the width and resolution of the image. By careful tuning of
these hyper parameters, the complexity and the model parameters are reduced.
On the other hand, in MobileNetv2, the architecture is represented in three
structures 1) depthwise separable convolutions, 2) linear bottleneck, and 3) Inverted
residual block. The depthwise separable convolution is divided into two parts. In the
first part, a filter is applied to each channel, and the second part involves 1x1 filters to
be applied to all channels, otherwise known as pointwise convolutions. In
MobileNetV2 (See Fig 3), pointwise convolutions and bottleneck layers are
combined. The main advantage of this operation is that when the ReLU activation
function is applied, lots of information is lost due to data dimensionality reduction. To
reduce the loss of information, a linear activation function is used in the bottleneck
layer. The inverted residual block is the expansion layer at the start of the block and
uses ReLU non-linearity. (See Fig 4)
6

Fig.4 This illustrates each block present in the MoblieNetV2 [8].

3.2 Proposed IoT module

Things and Internet

Things in IoT refer to multiple physical devices that are connected to the internet.
These things have specific features or properties like generating data, processing data,
transferring data, and unique identification. They are physical entities that have a
real-time presence and can be uniquely identified by other things. With the arrival of
super small micro-computers, including microprocessors and microcontrollers, it is
possible to bring the internet into any device irrespective of its size. When things are
connected to the internet, it unleashes a new dimension of possibility to innovate.

AI enabled IoT
IOT primarily consists of sensors that take input from their surroundings and
generate data. This data is of little utility unless it is preprocessed and analyzed
appropriately.AI can be integrated to analyze and predict the data generated by IoT,
So if data generated by IoT can be processed and used in the training of the AI model,
it can make IoT smarter. These intelligent IoT systems can help in automation without
any human intervention.

System Working
This system can be installed near the door or an entrance of a public place. As the
person comes near the range of the Passive Infrared motion sensor, the system gets
triggered. Then, a frame of the individual is captured using a camera module directly
connected to the Raspberry pi. The system proceeds to check for social distancing
using AI. If social distancing is violated, the person is alerted through voice
(speakers) and display messages (LCD). Otherwise, the system goes ahead with the
face mask detection.
For checking the presence of a mask, the system primarily locates the person's face
and crops it from the captured frame of the person closest to the door using the Face
Detector AI model. Later on, this cropped image is sent to the AI Mask Detection
model, which helps detect the mask. In the absence of a mask or if the mask is not
correctly worn, it alerts using voice and display outputs. Meanwhile, it also verifies
social distance by computing the distance between the identified individuals in the
frame using linear algebra. If the mask is worn appropriately, it continues to scan the
person's body temperature.
 7

To scan body temperature, an IR temperature camera consisting of multiple

integrated IR temperature sensors capable of scanning temperature from every pixel
of the image can be a viable solution. However, it suffers from the limitation of being
expensive. In real-time, the need to scan every pixel of an image becomes
unnecessary because we only need to scan a part of the image in most cases. Hence,
we can cut down the cost by employing a single external IR temperature sensor
instead of having multiple integrated sensors. This sensor is made to point at the
person's face.
Suppose, let us consider having a single fixed IR temperature sensor located at the
center of the camera. This sensor can scan the object's temperature located at the
center of the frame but misses out on all those who are not at the center or those who
are not of a certain height. Unlike IR temperature cameras, this typical IR temperature
sensor can scan temperature from a fixed point.
To resolve this issue, we employed servo motors and a 2D pan to orient the angle
of the temperature sensor towards the person's face and thus enabling the scanning of
the person's temperature without any miss. This reduces the cost incurred for sensors
by 95%.IR temperature sensor can scan the temperature only when the location
coordinates are specified. Therefore, AI can assist in locating the person's location
coordinates from the captured frame for scanning temperature. These coordinates help
us find the optimal angular orientation with respect to the x-axis and y-axis servo
motors in order to precisely point the temperature sensor at the person's face. These
calculated angles are set to the servo motors using a Pulse Width Modulation (PWM)
signal. At this point, the value obtained from the temperature sensor corresponds to
the person's body temperature. This whole operation takes less than a half-second to
complete.
In a nutshell, a person's face location is initially identified, the temperature sensor
is pointed towards the person's face using servo motors, and the body temperature is
scanned. Consequently, we unlock the door using a 12V solenoid valve. Parallelly,
LCD screen and speakers are used to communicate with the user.

Sensors and components

Raspberry Pi 4
Raspberry Pi is a small, robust, education-oriented, and credit card-sized computer
board. It is similar to a standard computer that runs on Linux. It uses a Broadcom
BCM2711 SoC with a 1.5 GHz 64-bit quad-core ARM Cortex-A72 processor,
on-board 802.11ac Wi-Fi, Bluetooth 5, full gigabit Ethernet, two USB 2.0 ports, two
USB 3.0 ports, and dual-monitor support via a pair of micro HDMI (HDMI Type D)
ports for up to 4K resolution. Most importantly, it has 40 GPIO pins that can be used
to connect with multiple other devices.
Resembling a typical computer, Raspberry Pi uses an operating system and delivers
the same user-friendly functionality. We employ the Linux distribution, Raspbian, for
its accessible, open-source, powerful, and compatible features for Raspberry Pi. There
are also a few non-Linux OS options available [11]. Raspberry pi is known for its
widespread application in various projects [12]. The drawback of using Raspberry Pi
in terms of power supply is that no external device can draw more than 100mA from
8

its USB ports [13]. Hence it is advised that all external components connected to
raspberry pi must be powered externally. In our system, raspberry pi controls all other
components and is responsible for processing the AI models within the system.

Raspberry camera v2
The credit-card-sized raspberry pi computer supports several accessories like a
camera module containing the Sony IMX219 sensor. This integration provides us with
the formatted raw data images that can be of value in applications like computer
vision, remote sensing, astronomy, improved image quality, and high dynamic range
imaging. This camera module can be directly connected to raspberry pi and can be
used autonomously without any configuration. We use a raspberry pi v2 camera to
capture frames of a person in raw format, which are then processed using image
processing techniques and sent to AI models for prediction.

GY-906 MLX90614ESF Temperature Sensor

For for our system, we utilized the MLX90614 infrared thermometer used
primarily for the purpose of non-contact temperature scanning. Features include a
built-in low noise amplifier, 17-bit Analog-Digital Converter, and a powerful DSP
module that account for high precision and accuracy. Portability, cost efficiency, and
ease of integration are some of its perks. Its temperature range spans from −40 °C to
125 °C for sensor temperature and −70 °C to 380 °C for object temperature [14], [15].
This sensor is merged with our system to scan the body temperature of the person.

Servo motors
Servomotors are popular for providing accurate angular precision by enabling the
user to rotate as per the given input [16]. The applied angular orientation remains
intact even in the absence of a power supply, unlike the working of a conventional DC
motor. Moreover, rotational progress can be controlled by a servo motor, whereas
only speed (RPM) can be controlled in the case of a DC motor. Its performance is
enhanced with its error sensing feedback functionality. We use two servo motors in
our system, one for each axis(x and y), to point the temperature sensor towards the
specified coordinates.

2 D Pan
2 Axis Pan Tilt Brackets are based on two axis pan and tilt mechanism for
mounting wireless or wired cameras and sensors on robot assembly. Panning, rolling,
and tilting are achieved by controlling Servo motors using PPM pulses. Suitable for
DIY panning and tilting application in projects. It supports a varied array of sensors
and measurement devices for different project applications. In our system, a 2 Axis
pan coupled with two servo motors helps to orient the temperature sensor towards the
person's face.

Motion detector
PIR motion sensor detects infrared radiation that lies in the wavelength range of
0.7 and 300 micrometers emitted from the human body. PIR sensor is widely used in
the security system to detect the motion of humans [17]. The normal human body
reflects IR at wavelengths ranging from 10 micrometers to 12 micrometers [18] [19].
 9

The infrared radiation of the human body PIR sensors is a passive electronic device
that detects motion by sensing infrared fluctuations [20]. It has three pins (gate, drain,
and source). After it has detected the IR radiation difference, a high is sent to the
signal pin. This helps our system sleep and save power while no one is around and
notifies the system immediately after someone approaches.

Joy Stick
An analog joystick, sometimes called a control stick joystick or thumb stick, is an
input device for a controller used for two-dimensional input. An analog joystick is
similar to two potentiometers, one for the vertical movement (Y-axis) and the other
for the horizontal movement (X-axis). The joystick also comes with a select switch.
The user can use this to change the settings of the system if needed.

16x2 LCD Module

16x2 LCD Module is a display that utilizes the light modulating properties of
liquid crystals to display pictures or text on a screen. It is used to provide required
information and interact with the user.

Speakers
3.5mm audio jack stereo speakers are used with raspberry pi to provide audio
output to the user.

Fan
The typical cooling fan used in desktop computers is used in this system to keep
raspberry pi and other components at normal temperature.

Bucks and boost converters

These are used to step up or step down the voltages as needed. Every component
needs a specific rated voltage to function properly. As we are using multiple
components, we require different levels of voltages. These required voltages can be
achieved using bucks and boosters.

Experimental Setup
T. The semantic circuit was initially simulated and later designed using easy EDA
software, and then all the components were soldered onto a perf board. Each IoT
component has a specific data transmission protocol. These protocols include Serial
Peripheral Interface (SPI), inter-integrated circuits (IIC), Universal asynchronous
receiver-transmitter (UART), serial transmission, and parallel transmission.
Depending on which protocol it follows, the circuit wiring and boiler code are written.
The casing is initially designed using solid works software and later laser-cut on
acrylic sheets. These acrylic sheets are bound together using L-shaped brackets. All
the electrical components can be fitted into the system in their specific slot. This
entire system can be installed on a wall using wall mount provisions on the system's
back panel.
10

4 Results and Discussions

The proposed system combines AI and IoT technologies and showcases their potential
in combating covid-19. For our training, the model is initialized with proper
initializes. The proposed classification model is trained with a batch size of 32 and a
learning rate of 1e-4 for 20 epochs. Our model got an accuracy of 99% with a
minimal loss of 0.071. The precision for with and without mask classes are 0.999 and
1.00, respectively, and an F1 score of 0.99 was obtained for both classes. The
validation accuracy and loss of the model(As shown in Fig 5).

Fig.5 Loss and accuracy of the model

The system was able to accurately scan each person's body temperature within 0.5
seconds, keeping in view the speed constraint associated with the servo motors. This
could be a pitfall that can compromise the performance of our system occasionally.
Being one of the smallest microprocessors, the raspberry pi has a performance
constraint that affects the capturing speed and data processing. The highest recorded
frames per second (FPS) of our system was 8. Compared to micro-controllers or older
versions of raspberry pi, the newer raspberry pi 4 consumes higher power (20 watts).
Hence, the battery life of our system is shorter than the desired level.

5 Conclusion
This paper introduced a novel approach to an AI-based covid screening system that is
both scalable and cost-effective. We achieved the automation of tasks like mask
detection, temperature scanning, and checking social distancing using AI and IoT
technologies. The Deep neural network module performed well in detecting faces
from captured images, and the MobileNetV2 proved to be a better alternative in
correctly classifying the images with an accuracy of 99%. It also displayed and gave
warning signs of those who did not meet the criteria. The system successfully granted
access for all the authorized people by displaying the appropriate message and
 11

opening the door for them. All the logs, including images, were stored locally for
further reference.

6 Limitations and Future scope

We can incorporate a Radio frequency identification (RFID) module into our system
to provide us with user authentication functionality. A web application can be
deployed to store and access images online that can be used for various purposes. The
entire casing of the system can be done using injection molding instead of laser-cut
acrylic sheets to increase durability. A more credible dataset with real-time images of
people wearing masks can increase the performance of our model. The use of a more
sophisticated microprocessor running from Windows operating systems like
Jetsonnano made by NVIDIA can be an upgrade to our system. The accuracy of our
system is compromised in the case of densely populated places or when the vicinity
range is greater than 5 meters. The installation of an external display screen can
improve the user interaction with the system.

References
1. M. M. Rahman, M. M. H. Manik, M. M. Islam, S. Mahmud and J. -H. Kim, "An
Automated System to Limit COVID-19 Using Facial Mask Detection in Smart City
Network," 2020 IEEE International IOT, Electronics and Mechatronics Conference
(IEMTRONICS), 2020, pp. 1-5, doi: 10.1109/IEMTRONICS51293.2020.9216386.
2. S. A. Sanjaya and S. Adi Rakhmawan, "Face Mask Detection Using MobileNetV2 in The
Era of COVID-19 Pandemic," 2020 International Conference on Data Analytics for
Business and Industry: Way Towards a Sustainable Economy (ICDABI), 2020, pp. 1-5,
doi: 10.1109/ICDABI51230.2020.9325631.
3. Nagrath, P., Jain, R., Madan, A., Arora, R., Kataria, P., & Hemanth, J. (2021). SSDMNV2:
A real time DNN-based face mask detection system using single shot multibox detector
and MobileNetV2. Sustainable cities and society, 66, 102692.
https://doi.org/10.1016/j.scs.2020.102692.
4. Singh, R. P., Javaid, M., Haleem, A., & Suman, R. (2020). Internet of things (IoT)
applications to fight against COVID-19 pandemic. Diabetes & metabolic syndrome, 14(4),
521–524. https://doi.org/10.1016/j.dsx.2020.04.041.
5. Mukhtar, H., Rubaiee, S., Krichen, M., & Alroobaea, R. (2021). An IoT Framework for
Screening of COVID-19 Using Real-Time Data from Wearable Sensors. International
journal of environmental research and public health, 18(8), 4022.
https://doi.org/10.3390/ijerph18084022
6. Abdulrazaq, Assoc. Prof. Dr. Mohammed & Zuhriyah, Halimatuz & Al-Zubaidi, Salah &
Karim, Sairah & Ramli, Rusyaizila & Yusuf, Eddy. (2020). NOVEL COVID-19
DETECTION AND DIAGNOSIS SYSTEM USING IOT BASED SMART HELMET.
International Journal of Psychosocial Rehabilitation. 24. 2296-2303.
10.37200/IJPR/V24I7/PR270221.
7. dS. Karmore, R. Bodhe, F. Al-Turjman, R. L. Kumar and S. Pillai, "IoT Based Humanoid
Software for Identification and Diagnosis of Covid-19 Suspects," in IEEE Sensors Journal,
DOI: 10.1109/JSEN.2020.3030905.
8. Sandler, Mark, et al. "Mobilenetv2: Inverted residuals and linear bottlenecks." Proceedings
of the IEEE conference on computer vision and pattern recognition. 2018.
12

9. Howard, Andrew G., et al. "Mobilenets: Efficient convolutional neural networks for
mobile vision applications." arXiv preprint arXiv:1704.04861 (2017).
10. Howard, Andrew, et al. "Searching for mobilenetv3." Proceedings of the IEEE/CVF
International Conference on Computer Vision. 2019.
11. M. Richardson and S. Wallace, Getting started with Raspberry Pi, O’Reilly, USA,
2013.
12. Maksimovic, Mirjana & Vujovic, Vladimir & Davidović, Nikola & Milosevic, Vladimir
& Perisic, Branko. (2014). Raspberry Pi as Internet of Things hardware: Performances and
Constraints.
13. M. Schmidt, Raspberry Pi – A Quick Start Guide, The Pragmatic Bookshelf, 2013.
14. T. Pam and L. T. H. D. N, “A Product Line of Diodes Incorporated FILTERLESS 3W
CLASS-D STEREO AUDIO AMPLIFIER Description Pin Assignments Features A
Product Line of Diodes Incorporated PAM8403 Pin Descriptions Functional Block
Diagram,” no. November, pp. 1–11, 2012.
15. P. I. N. Assignment and O. Information, “Trickle Charge Timekeeping Chip,” pp. 1–13.
https://docs.rs-online.com/46c9/0900766b80025c86.pdf.
16. J. Chen, X. Zou and F. Wang, "Research and Design of DC Servo Motor Position Control
System Based on LabView," 2010 International Conference on E-Product E-Service and
E-Entertainment, 2010, pp. 1-5, doi: 10.1109/ICEEE.2010.5660968.
17. L. Herbert, Berman, Altos Hills Los , and Calif, "Infrared Intrusion Detector System,"
3,703,718, Nov. 21, 1972.
18. S. Yuvaraj Prof. and Ramesh S., "Improved Response Time on Safety Mechanism Based
on PIR sensor," International Journal of Emerging Technology and Advanced Engineering,
vol. 2, no. 4, April 2012.
19. Anonymus (2007, March 13), “The Electromagnetic Spectrum” [online]
http://science.hq.nasa.gov/kids/imagers/ems/infrared.html
20. David W. Crick and West Molesey, "Passive Infrared intruder Detector System,"
4,242,6690, December 30, 1980.