ARTIFICIAL INTELLIGENCE BASED SMART NAVIGATION SYSTEM FOR BLIND PEOPLE

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 07 | July 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 2295
ARTIFICIAL INTELLIGENCE BASED SMART NAVIGATION SYSTEM
FOR BLIND PEOPLE
Dr.S. Mary Joans*, Aishwarya R1, Meenu Sam2, Sree Ranjane C3, Subiksha S4
Department of Electronics and Communication Engineering,
Velammal Engineering College, Chennai-600066, Tamil Nadu, India.
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - There are a huge number of blind and visually
impaired individuals in this world who consistentlyneedsome
assistance. These blind people find it difficult to go outside
their homes without any help. Artificial Intelligence Based
Smart Navigation System is a gadget intended to help guide
the visually impaired by recognizing objects and depictingthe
data to them as an audio signal. This diminishes human
exertion and gives better comprehension of the surroundings.
Besides it likewise gives a chance for the blind individuals to
move from one spot then onto the next without needing any
help from others. The gadget can likewisebeutilizedinoldage
homes where aged individuals experience issues in their
everyday exercises because of diminishedvision. Inthiswork, a
framework is carried out that navigates the individual along
his/her way. The object before the visually impaired can be
recognized using imagerecognition withArtificialintelligence.
The data about the object type before the individual can be
conveyed by means of earphones / headphones. Thiswillassist
the individual in walking without any trouble.
Key Words: Artificial Intelligence, Visually Impaired, Blind,
Obstacle Detection, Recognition, Accuracy.
1. INTRODUCTION
As indicated by World Health Organization (WHO),
there are over 1.3 billion individuals who are physically
impaired across the globe, out of which about 36 million
individuals are blind. India being the second biggest
populated country on the planet, contributes about 30% of
the total blind population. In spite of the fact that enough
missions are being directed to treat these individuals, it has
been hard to source every one of the prerequisites. It istruly
challenging for blind individuals to roam openly in the
outside world. Individuals with visual inabilityhavea steady
need of help with their regular routines. It can go from
depending on others for help for basic needs to utilizing
various Electronic Travel Aid (ETA) assistive gadgets
whenever needed. They also usewhitecanesandguidedogs.
However, the limitation on the allowance of guide dogs in
specific places and the short scope of white canes fill in as
some of the disadvantages. Most of the ETAs have the
disadvantages of being unreasonably expensive and
wasteful. Keeping a tab on the different downsides of the
conventional and normal techniques for helping thevisually
impaired and the blind individuals, we chose to create and
plan our framework with a novel approach.
This era of brilliant advancements like Artificial
Intelligence and Machine Learning implanted with
appropriate equipment has life of visually impaired
individuals simple and more secure. In the recent years,
there is a wide-scale use of Artificial Intelligence going from
scientific research, automated vehicles,national defenceand
space exploration. It is the need and obligation of the
analysts to create such equipment for blind individuals to
stroll around any region with the guide of smart navigation
system that is built and most arising advancements, for
example, Artificial Intelligence and Machine learning. AI has
acquired big importance because of the huge amountofdata
and simplicity of calculation. Utilizing AI it is feasible to
make these individuals' life a lot easier. The objective is to
give an "optional sight" until they have an adequate number
of assets needed to treat them. Individuals with untreatable
visual impairment can utilize this to make their ordinary
tasks less difficult.
In this paper, we propose a novel smart navigation
system for blind and the visually impaired individuals.Some
of the unique features of the system include:
1) A real-time and a camera-based system that is
simple in design and also reduces the cost by
minimizing the number of sensors.
2) A compact and a low-power design with an
integrated voice alert system for outdoor and
indoor navigation.
3) Processing of complex algorithm with a low-end
configuration.
2. RELATED EARLIER WORKS
V.Kunta et al.[1] proposed a system that makes use
of the Internet of Things (IoT) to create a bridgebetweenthe
environment and the blind. Many sensors are used to detect
obstacles, damp floors, and staircases, among many other
things. The presented prototype is a basic and inexpensive
smart blind stick. It is fitted with a variety of IoT modules
and sensors.

A.A.Diaz Toro et al.[2] presented the methodology
that can be used to build a vision-based wearable system
which will assist visually impaired people with navigationin
indoor atmosphere that is new to them. The proposed
system helps the person in “purposeful navigation”. The
system detects obstacles,walkablespacesandothersobjects
like computers, doors, staircases, chairs etc.
M.M.Soto-Cordova et al.[3] presented the design as
well as the implementation of a helping aid for the blind.It is
implemented using Arduino, ultrasonic sensors and some
warning devices for obstacle detection and intimationbased
on the distance of the obstacle.
N.Loganathan et al.[4] presented a solution by
implementing an ultrasonic sensor in the cane. The device
can detect and perceive the obstacles at a range of 4 meters.
The IR device will detect the nearby obstacles that approach
the blind. With the help of a buzzer, the radio frequency
transmitter and receiver are used to findthe preciselocation
of the stick.
J.Ai et al.[5] proposed a device that first acquires
image information with a camera before converting the
image to text with image captioning technology. Finally, the
user is fed back with the text sequence using a voice signal.
D.Bal et al.[6] introduced a wearable system for the
blind people that can be used for indoor navigation. It is a
lightweight, portable and a user friendly solution which can
also be mounted on a jacket. The system uses Raspberry Pi,
vibration motors, a camera module,threeultrasonic sensors,
emergency button, gyroscope and the android application
‘Blynk’.
M.A.Khan et al.[7] designed an obstacle avoidance
architecture with a camera and sensors, as well as advanced
image processing algorithms to detect the objects. The
distance between the obstacle and the user is determined
using an ultrasonic sensor and a camera. The system
contains an image - text converter and audio feedback, as
well as an integrated reading assistance.
S.Dev et al.[8] demonstrated a smart device made
using a Raspberry Pi that can perform tasks like object
detection, navigation and gives audio feedback to notify the
user.
D.P.Khairnar et al.[9] proposed a system that
combines smart gloves with a smartphone application. The
smart glove detects and avoids obstacles, as well asallowing
visually impaired persons to recognize their surroundings.
Various things in the neighbourhood are detected using
smartphone-based obstacle and object detection.
Devi et al.[10] proposed a system that performs
existing object observation and discovers the actions. The
nearby location can be captured and given as a voice
command. The proposed system effectively performs live
observation of visually impairedandblind peopleandallows
them to move anywhere without any help.
S. Barathi Kanna et al.[11] proposed a finding that
allows visually impaired people and blind people to
"communicate" with their surroundings in the Internet of
Things (IoT) world. This prototype comprisesanESP8266,a
power source for the developmentboardandcoinmotors,as
well as a smartphone App, making it more accessible to
visually impaired people.
K. Patil et al.[12] developeda conceptfora wearable
device with a virtual assistant that would allow visually
impaired and blind persons to accomplish simple tasks
without assistance. The system is designed to providevoice-
over assistance to visually impaired people who need help
with their day to day life tasks.
3. PROPOSED SYSTEM
The Proposed system consists of a camera, AI
processing, controller and voice alert. The ultimate goal of
this smart navigation system is to detect the obstacles
coming in front of the visually impaired person and to
inform them about the object. AI processing is used to train
the object images in the controller. If any objects/obstacles
approach the blind person, then the controller alerts the
person by a voice message. This can make the blind person
more cautious and thereby lowers the possibilities of
accidents. An automatic switch enabled with a voice
assistant is also added to guide them in their privatespace.A
camera is used to capture the indoor and outdoor images
which help the smart navigation system to detect the
obstacle.
The below flowcharts show the proposed system flowchart
and the block diagram.
Fig-1: Proposed block diagram

Fig.-2: Block diagram of AI processing
Fig-3: Proposed system flowchart
The flowchart of the proposed system is shown in
figure 3. The hardware part consists of an ultrasonic sensor,
camera, voice alert, and AI processing-based controller unit.
These trained data have described the use of ultrasonic
sensors integrating it with a controller for detecting the
obstacles. The ultrasonic starts sending the signals with
minimum delay and after that, the signal returns as an echo
to the sensor receiver, and the controller estimates the time
it takes to get the signal back from the sensor. Using thetime
taken, the distance of the object is calculated and is
converted into speech and the feedback is given through
speakers or headphones. The blind person will then be
stopped from colliding with any object.
This assistive framework is built based on the idea
of the Electronic Travel Aid (ETA), which plans to help the
visually impaired and the blind people in strolling while at
the same time staying away from obstacles.Letus considera
situation in which a chair is present in front of the visually
impaired person. The person wants to sit and take a rest on
the chair. In this case, the chair is not just an obstacle but a
useful object. While using the Obstacle detection system, the
user may not know what kind of obstacle is present in front
of them but has to confirm it on their own. But on the other
hand, if the user uses the Object recognition system, it helps
to determine the object to be a chair and the user can claima
benefit. It is important to construct an assistive framework
to detect and determine the objects around a visually
impaired user.
4. SOFTWARE DESCRIPTION
4.1 ARD Object Tracker
This is an Android application that uses OpenCV
libraries for computer vision detection.The ARD object
tracker application can detect and track various types of
objects from your mobile phone’s camera such as lines,
colour blobs, circles, rectangles and also people. Detected
object types and screen positions can then be sent to a
Bluetooth receiver device such as HC-05.
If using an appropriate micro-controller e.g. Arduino or
Raspberry Pi users can analyze the detected objects for
further projects.
Key Application Features:
1. Colour Blob Detect and Track
2. Circle Detect and Track
3. Line Detect
4. People Detect and Track
5. Rectangle detection.
6. Send detected object parameters wirelessly over
Bluetooth.
 Setting up Bluetooth:
To connect an appropriate Bluetooth receiver, click the
"Connect Bluetooth" button. If there is no button displayed
you will need to double or single click the preview screen to
bring up the options. The list shown provides the existing
paired devices for your phone. If there are no items on the
list, you will need to first pair your devicewithintheAndroid
Bluetooth settings options. Clicking on an existing paired
device will attempt a connection to that device. If successful
you will be taken back to the camera preview screen. Any
errors will be displayed if not successful.

 Configuration Settings:
There are several configurationsettingsthatcontrol howthe
OpenCV computer vision libraries detect objects. These can
be set by first clicking the "Settings" button. If there is no
button showing on the preview camera screen, users will
need to double or single click the preview screen to access
the settings page.
 Video Scale:
This controls the image size that will be used for all image
processing operations. A scalefactorof1will usethephone’s
default image size capped to a maximum of 1280x720. If
your phone’s CPU is not fast, it is recommended to set this
value to 2. Higher scale factors will result in faster image
processing operations butcanpotentiallyresultininstability
or no object detections.
 Tracked Object Persistence:
Each tracked object is given a unique id. Tracked objects
however can be missed on certain image frame updates and
consequently re-appear on subsequent frames. To avoid
issuing new ids, each tracked object is given a life
persistence value. Increasing this configuration parameter
will keep objects being tracked longer in the memory
buffers.
 BT Serial Tx Buffer Size:
This configuration parameter controls how many detected
object items are pushed onto the internal communication
buffer for Bluetooth communication. The internal
communication buffer is sampled 100 times a second
(10ms). This should be sufficient to clear the buffer.
 Bluetooth Data Transmit Formats:
All data communication is sent as ASCII text in the following
format:
"Object Type":"ID":"XPos","YPos","Width","Height"
4.2 Adaboost Training Algorithm
AdaBoost is an iterative algorithm that joins a new
weak classifier in each round until a desired lowerrorrateis
reached. A weight is assigned to each training sample,
indicating the possibility of it being included in the training
set by a classifier. If a sample point is successfully classified,
its weight will be reduced when building the next training
set. On the other hand, if a sample point is misclassified, its
weight will be increased. In this implementation, each
sample has the same initial weight, thus we will choose
sample points based on these weights for the kth iteration
operation, and then train the classifier. The weight of the
sample that is incorrectly classified can then be increased,
while the weight of the sample that is successfully classified
can be reduced, based on this classifier.Themodifiedsample
set weights are then utilized to train the next classifier. This
is how the entire training procedure is carried out.
Description of AdaBoost algorithm :
(i) Sample: Given the following training sample set:
(x1, y1), (x2, y2),..., (xn, yn),
where xi ∈ X, yi ∈ Y = {−1, +1}. (1)
(ii) Initialize the sample's weights:
 For positive samples, use the formula
W1(i) = 1/2m (2)
where m represents the total number of positive
samples.
 For negative samples, use the formula
W1(i) = 1/2l (3)
where l represents the total number of negative
samples.
(iii) Weak classifiers are trained by :
Circulate t = 1,..., T: where T represents training
rounds.
(iv) (4)
where α t represents the weight of weak
classifiers and t represents the classifier's lowest
error rate at the current level.
(v) If the sample is correctly identified, revise the
weights with
Wt+1(i) = Wt(i) * exp ( (5)
If the sample is incorrectly classified,
Wt+1(i) = Wt(i) * exp ( (6)
(vi) Final strong classifier = (Sum of weak classifiers)
* (Individual weights of weak classifiers).

Fig-4: How weak classifiers work
Advantages of AdaBoost algorithm
 The AdaBoost algorithm offers the framework for
multiple approachestogeneratesub-classifiers,and
the AdaBoost algorithm is a simple classifier with
excellent accuracy.
 The estimated findings are simple to understand,
and constructing a weak classifier is simple.
 There is no necessity for feature selection
 There is no threat of over fitting.
Improved Adaboost algorithm
The upper bound of the least error rate on the
training set is used as a criterion for setting parameters in
the traditional AdaBoost approach. However, this does not
guarantee that the error rate will be reduced. As a result,the
weak classifier is no longer picked as the lowest errorratein
each training session in an upgraded AdaBoostapproach.To
tackle this problem, a weighting parameter of the weak
classifier can be added to boost the performance of the
improved AdaBoost algorithm. Furthermore, if the training
set contains difficult samples and noise, the weak classifier's
capability to generalize will deteriorate as the number of
repetitions grows. "Degeneration" is the term for this
condition. The weight expansion must be limited in this
circumstance.
5. SIMULATION RESULTS
MatLab was used for carrying out simulation. The
first test was carried out using 600 training sets which
includes 400 negative and 200 positive samples.
Table-1: Accuracy and Time of 600 training sets
Chart-1: Number of weak classifiers vs Accuracy
Chart-2: Number of weak classifiers vs cost time
From chart 1, it is understood that in traditional
AdaBoost algorithm the accuracy is represented by theblue
line whereas in improved AdaBoost algorithm it is
represented by the green line. For 100 weak classifiers, the
accuracy of traditional AdaBoost algorithm is 82.50%
whereas for the improved AdaBoost algorithm is 83.50%.
The cost time is observed as 2.5980 seconds for the
traditional AdaBoost algorithm and 2.6180 seconds for the
improved AdaBoost algorithm. It is also seen that the cost
time for both the algorithms is similar.
It is also seen from the graph that the accuracy of the
improved AdaBoost algorithm is increasingswiftlywhereas
for traditional AdaBoost algorithm the accuracy remains
unchanged for the number of weak classifiers 200 – 500.
When it is 500, the greatest contrast in the accuracy is seen.
The accuracy of the traditional AdaBoost algorithm is seen
to be 4% less than the improved AdaBoost algorithm. Incan
be observed that both the algorithm have almost the same
cost time. In the traditional AdaBoost algorithm, about 700
weak classifiers have to be trained and the cost time is
16.4470seconds, however in the improved AdaBoost
algorithm only 400 weak classifiers need to be trained and

the cost time is 8.1710seconds, hence it saves about
8seconds.
Fig-5:Adaboost Training and detection process
6. PERFORMANCE ANALYSIS
Thebelowtableshowstheperformancecomparison
of different AI algorithms used for different objects. The
classifier based on AdaBoostproducesmoreaccurateresults
than the other classifiers.
Table-2: Performance comparison of different AI
algorithms
The recognitionaccuracy comparison is shown inbargraph
in the following figure.
Chart-3: Recognition accuracy comparison
7. HARDWARE DESCRIPTION
 Arduino Uno controller
It is a microcontroller basedontheATmega328that
has 20 digital input/output pins, ICSP header, USB port, a
power jack, reset button and a 16 MHz resonator.
Fig-6: Arduino Uno Controller
 Camera
A mobile camera or a webcam is used to record the
images of the obstacles that approach the path of the blind
person. For image detection, the clustering technique is
utilized, with the image of the obstacles as input. To identify
these inputs, they are compared to images recorded in the
database. Then the output is given in the form of an audio
signal if the input image matches with the database.
 Ultrasonic sensor
It is an electronic devicethat hasa transducerwhich
emits and receives ultrasonic sound waves to detect the
distance to an object and provides information about an
item’s proximity.
S.no Algorithm
Recognition
accuracy
1 Random forest 89
2 Support vector machine 92
3 Neural network 92.3
4 Ada boost 95.6

Fig-7: Ultrasonic sensor
 DF Player mini Mp3 Player Module
The dfplayer mini is an easy-to-use, stable, reliable
and compact mp3 module that can connect to the speaker
directly. It integrates with MP3, WAV, and WMA hard-
decision decoding. Tf cards with FAT16 and FAT32 file
systems are also supported by this module.
Fig-8: DF Player mini Mp3 Player Module
 HC-05 Bluetooth module
It is used to connect Arduino to other devices. The
HC-05 is used to provide a wireless serial connection that is
transparent. It operates on the 2.45GHz frequency band and
requires a power source of 4-6V. The data transfer rate
varies from 1Mbps to 10Mbps and is within a 10-meter
range.
Fig-9: HC-05 Bluetooth module
 Battery
The smart navigation system's electrical power is
supplied by a battery. In order to process the equipment
connected to the smart navigation system, a 9 volt battery is
used.
Fig-10: Battery
 Speaker
The speaker module that is used has a good
performance and is generally used for all types of audio
projects. 8 Ohm speakersconsumelesspowercomparedto4
Ohm speakers. They also emit less heat from the voice coil
and deliver high-quality sound and can handle about 200
watts of power.
Fig-11: Speaker
8. IMPLEMENTATION RESULTS
The proposed system is implemented using an
Arduino controller. The circuit connection and overall
implementation of the proposed model are depicted in the
figure below.
Fig-12: Overall implementation
9. CONCLUSION
The blind population may step inorout withoutany
help while utilizing this smart navigation system as it has
various functions including image processing, obstacle
detection, and text to speech. Furthermore, our system does
not need any input from the user, as it consists of ultrasonic
sensors and camera modules thatsenseobstaclesaround the
user and provide feedback to the user with the assistance of

an audio device connected to the module. The system
continually operates so that the visually disabled can get
updates on the challenges at any point on the way.
Consequently, with the aid of our project, blind peoplecould
perform their daily tasks comfortably without any trouble.
10. REFERENCES
[1] V. Kunta, C. Tuniki and U. Sairam, "Multi-Functional
Blind Stick for Visually Impaired People," 2020 5th
International Conferenceon CommunicationandElectronics
Systems (ICCES), 2020, pp. 895-899.
[2] A. A. Diaz Toro, S. E. Campaña Bastidas and E. F.
Caicedo Bravo, "MethodologytoBuilda WearableSystem for
Assisting Blind People in Purposeful Navigation," 2020 3rd
International Conference on Information and Computer
Technologies (ICICT), 2020, pp. 205-212.
[3] M. M. Soto-Cordova,F.Criollo-Sánchez,C.Mosquera-
Sánchez and A. Mujaico-Mariano, "Prototype of an Audible
Tool for Blind Based on Microcontroller," 2020 IEEE
Colombian Conference on Communications and Computing
(COLCOM), 2020, pp. 1-4.
[4] N. Loganathan, K. Lakshmi, N. Chandrasekaran,S.R.
Cibisakaravarthi, R. H. Priyanga and K. H. Varthini, "Smart
Stick for Blind People," 2020 6th International Conference
on Advanced Computing and Communication Systems
(ICACCS), 2020, pp. 65-67.
[5] J. Ai et al., "Wearable Visually Assistive Device for
Blind People to Appreciate Real-world Scene and Screen
Image," 2020 IEEE International Conference on Visual
Communications and Image Processing (VCIP), 2020, pp.
258-258.
[6] D. Bal, M. M. Islam Tusher, M. Rahman and M. S.
Rahman Saymon, "NAVIX: AWearableNavigation System for
Visually Impaired Persons," 2020 2nd International
Conference on Sustainable Technologies for Industry 4.0
(STI), 2020, pp. 1-4.
[7] M. A. Khan, P. Paul, M. Rashid, M. Hossain and M. A.
R. Ahad, "An AI-Based Visual Aid With Integrated Reading
Assistant for the Completely Blind," in IEEE Transactions on
Human-Machine Systems, vol. 50, no. 6, pp. 507-517, Dec.
2020.
[8] S. Dev, S. Jaiswal, Y. Kokamkar, K. B. Deshpande and
K. Upadhyaya, "Voice Based Smart Assistive Device for the
Visually Challenged," 2020 International Conference on
Convergence to Digital World -QuoVadis(ICCDW),2020, pp.
1-5.
[9] D. P. Khairnar, R. B. Karad, A. Kapse, G. Kale and P.
Jadhav, "PARTHA: A Visually Impaired Assistance System,"
2020 3rd International Conference on Communication
System, Computing and IT Applications (CSCITA), 2020, pp.
32-37.
[10] Devi, M. J. Therese and R. S. Ganesh, "Smart
Navigation GuidanceSystem forVisuallyChallengedPeople,"
2020 International Conference on Smart Electronics and
Communication (ICOSEC), 2020, pp. 615-619.
[11] S. Barathi Kanna, T. R. Ganesh Kumar, C. Niranjan,S.
Prashanth, J. Rolant Gini and M. E. Harikumar, "Low Cost
Smart Navigation System for the Blind," 2021 7th
International Conference on Advanced Computing and
Communication Systems (ICACCS), 2021, pp. 466-471.
[12] K. Patil, A. Kharat, P. Chaudhary, S. Bidgar and R.
Gavhane, "Guidance System for Visually Impaired People,"
2021 International Conference on Artificial Intelligence and
Smart Systems (ICAIS), 2021, pp. 988-993.
[13] S. Khan, S. Nazir and H. U. Khan, "Analysis of
Navigation Assistants for Blind and Visually Impaired
People: A Systematic Review," in IEEE Access, vol. 9, pp.
26712-26734, 2021.
[14] S. Malkan, S. Dsouza, S. Dasmohapatra,M.JainandA.
Joshi, "Navigation using Object Detection and Depth Sensing
for Blind People,"2021IEEE4thInternational Conference on
Computing, Power and Communication Technologies
(GUCON), 2021, pp. 1-7.
[15] N. Nagenthiran, P. Priyanha, S. Nirosha, J. Vivek, H.
De Silva and D. Sriyaratna, "Machine Learning-Based Smart
Shopping for Visually Impaired," 2021 3rd International
Conference on Advancements in Computing (ICAC), 2021,
pp. 473-478.
[16] N. Bhandary, K. Siddique, A. Adnani,S.Sengupta and
S. Patil, "AI-Based Navigator for hassle-free navigation for
visually impaired," 2021 International Conference on
Computer CommunicationandInformatics(ICCCI),2021, pp.
1-6.
[17] D. M. L. V. Dissanayake, R. G. M. D. R. P. Rajapaksha,
U. P. Prabhashawara, S. A. D. S. P. Solanga and J. A. D. C.
Anuradha Jayakody, "Guide-Me: Voice authenticated indoor
user guidance system," 2021 IEEE 12th Annual Ubiquitous
Computing,Electronics&MobileCommunicationConference
(UEMCON), 2021, pp. 0509-0514.

ARTIFICIAL INTELLIGENCE BASED SMART NAVIGATION SYSTEM FOR BLIND PEOPLE

More Related Content

Similar to ARTIFICIAL INTELLIGENCE BASED SMART NAVIGATION SYSTEM FOR BLIND PEOPLE (20)

More from IRJET Journal (20)

Recently uploaded (20)

ARTIFICIAL INTELLIGENCE BASED SMART NAVIGATION SYSTEM FOR BLIND PEOPLE