AN INTELLIGENT SYSTEM FOR THE ENHANCEMENT OF VISUALLY IMPAIRED NAVIGATION AND DISASTER ASSISTANCE USING GEO-BASED POSITIONING AND MACHINE LEARNING

International Journal of Grid Computing & Applications (IJGCA) Vol.12, No.1/2/3, September 2021
DOI:10.5121/ijgca.2021.12301 1
AN INTELLIGENT SYSTEM FOR THE
ENHANCEMENT OF VISUALLY IMPAIRED
NAVIGATION AND DISASTER ASSISTANCE USING
GEO-BASED POSITIONING AND
MACHINE LEARNING
Wenhua Liang1
, Ishmael Rico2
and Yu Sun3
1
St.Margaret’s Episcopal School, San Juan Capistrano, CA 92675
2
University of California, Berkeley, CA, 94720
3
California State Polytechnic University, Pomona, CA, 91768
ABSTRACT
Technological advancement has brought the masses unprecedented convenience, but unnoticed by many, a
population neglected through the age of technology has been the visually impaired population. The visually
impaired population has grown through ages with as much desire as everyone else to adventure but lack
the confidence and support to do so. Time has transported society to a new phase condensed in big data,
but to the visually impaired population, this quick-pace living lifestyle, along with the unpredictable nature
of natural disaster and COVID-19 pandemic, has dropped them deeper into a feeling of disconnection from
the society. Our application uses the global positioning system to support the visually impaired in
independent navigation, alerts them in face of natural disasters, and reminds them to sanitize their devices
during the COVID-19 pandemic.
KEYWORDS
Geo-based navigation assistance, machine learning, mobile computing
1. INTRODUCTION
The advancement of technology has long been a significant accomplishment or representation of
progress made for the convenience of the entire human race. However, as technologies continue
to evolve for the fancy and quick-paced future that we all anticipate to witness, inconveniences
are left behind to some neglected populations like the elderly population and the population with
disabilities. Indeed, while technologies have contributed to the well-being of the mass and
revolutionized the future, it would be unfair to ignore how it has also widened the technological
generation gap, where the elderly population falls behind to adopt to new technologies, and the
capacity-ability gap, where the population with disabilities feel challenged in face of technologies
[1]. According to the Centers for Disease Control and Prevention’s Disability impacts All of Us
infographic, 61 million adults in the United States live with some type of disability, which is, in
fact, 1 in every 4 adults in the United States [2]. Among them, about 3 million experience
blindness and difficulty with seeing. Even worse, as technologies enhance in this rapidly evolving
world, problems have, otherwise, been introduced to the visually impaired population, and a list
of such includes difficulty with navigation, communication, and the maintenance of a sense of

2
independence by the visually impaired population in the ever-changing city. A research
conducted by Yan Wu’s team through questionnaires shows that while as much as 77% of
visually impaired survey participants appreciate the value provided by technologies, only 23 % of
responders own smartphones (due to restricted accessibility and insufficient financial supports)
compared to the 76% of sighted UK adults who own their own smartphones [4].
This clearly indicates that despite the immense technological advancement in this society, the
visually impaired population experience inadequately from this advancement. In fact, many lack
experience in even the very least of what are known to be successful technologies. Therefore, to
respond to the World Health Organization’s call to leave no one behind, our application aims to
guide the visually impaired population toward independence [3].
Some common systems implemented for the guidance of visually impaired in navigation are the
mobility tools like white canes and the guide dogs. The visually impaired people usually use the
white cane to feel the changes on the surface of the pathway ahead of their steps. By swinging the
cane from one side to another in rhythm with their feet, the users will be able to find and be
aware of the obstructions ahead of them [5]. However, the effectiveness of the cane can be
restricted when the cane breaks during a navigation, when poor weather or unfamiliar landmarks
negatively influence its performance and, most importantly, when the person has to cross an
intersection, considering that the white cane remains ineffective toward detecting vehicles in
motion. In contrast, the guide dog, a trained pet for leading the visually impaired around
obstacles, is aware of the danger imposed by the speeding vehicles and is thus a safer choice
when it comes to crossing intersections. Indeed, guide dog is preferred by many for its mobility
and companionship [6]. However, a guide dog can be an unconvincing choice for many because
it requires additional cares, training, and expenses on goods to feed it. While guide dog and white
cane are the most common systems employed by the visually impaired, they only provide
assistance in independent navigation but lack assistance in other forms. Another proposed system
that has helped the visually impaired to achieve independence is the subscription to real-time
agents in needs of assistance. The idea behind this is that the visually impaired population can
receive assistance immediately through a touch of a button or a call through applications on
smartphone or laptop. While this definitely provides more forms of assistance to the users other
than on independent navigation, the expenses are usually higher, and the assistance would not be
there to be accompanied with the user twenty-four-seven. In fact, some platforms would not work
without internet access, providing inconveniences and restrictions in navigation when the user
travels to locations with no internet and thus access to support. Electronic glasses is another
proposed system to help with not only the navigation of the user but in many other forms. In fact,
its principle builds upon the capturing of real-time images through a camera and the displaying of
these motions to the users’ sight, which basically allows the users to live in a world with clear
vision [7]. However, this piece of device is costly and not covered by medical insurances, so
though many hope to experience vision with such advanced technology, the majority of them
cannot afford it. Moreover, this device only works for those who are not completely blind, as it
functions upon enhancing one’s vision by displaying processed images.
In this paper, we follow the same line of research by assisting the visually impaired population to
achieve a sense of independence through integration of software, hardware, and machine-
learning. We have three goals to fulfill through this application:
1. Save data of the number of obstructions detected by the device in a one hour walk done by
a visually impaired person (frequency) with its corresponding longitude and latitude into a
database. With such database and through machine-learning, we will be able to predict
frequency at similar categorizes of location using the characteristics of the specific location

3
and to understand the relationship between frequency and different interior layout better for
designing the most traveling-friendly layout for the visually- impaired in the future.
2. Use frequency, longitude, and latitude to determine whether the user is indoor (in an
enclosed space) or outdoor (at an open space) so that instructions can be sent to the users
who are at an unsafe type of location, either an enclosed space or an open space, during the
specific natural disaster.
3. When reached a specific frequency, the user will be alerted to sanitize the device to avoid
potential contact with Covid-19.
Our method is inspired by our aims to understand what’s best for the visually impaired
population but to also support this community during natural disasters and the COVID-19
pandemic. There are some significant features on our application. First, our application is voiced
by computer instead of human agents, allowing for a greater scale of flexibility and accessibility
everywhere at any time. Second, our application covers functions outside of independent
navigation like an alerting system during natural disasters, which contributes toward the
independence of the visually impaired through different aspects of their life. Third, as a
technology itself, our device acts as a potential bridge to close the capacity-ability gap between
the visually impaired population and the society, and it also makes the positivity within the
advancement of technology apparent to the visually impaired population. Therefore, we believe
that this application will connect the visually impaired population closer to the society and
encourage them to be more independent and confident.
Experiments are performed to calculate and compare accuracies for determining the most
appropriate machine learning model through the implementation of support vector learning (svm)
and regression models. In the experiment, we tested different models through adjusting the
regression model, polynomial parameters, and inputted data sets.
The rest of the paper is arranged as follows: Section 2 presents the essential background for this
study. Section 3 provides the details on the challenges that the visually impaired population
encounters; Section 4 proposes solutions in response to the challenges mentioned in Section 3;
Section 5 discusses the relevant details about our process of experimentation, following by a
presentation on related work in Section 6. Finally, Section 7 offers the conclusion remarks, as
well as pointing out the future work of this project.
2. BACKGROUND STUDY
The section below illustrates current understanding and background information on cognitive
mapping without visual experience and machine learning algorithms used in this study.
2.1. Cognitive Mapping without Visual Experience
It is most essential to have a comprehensive understanding on cognitive mapping without visual
experience. The visual system has been concluded to be the most efficient in the obtention of
spatial information for its ability to perform foveation along with peripheral vision - which haptic
exploration does not support - and its outstanding performance in localization and object
identification, which audial exploration does not support. That is, due to the proposal of such
claim, questions have been raised on whether the visually impaired population is capable of
obtaining a spatial understanding without experience with visual cues. Three major theories
emerged and grew into debates as deficient theory, inefficiency theory, and the difference theory.
The deficiency theory suggests that the lack of visual experience leads to the consequence of a

4
total deficiency in spatial experience. On the other hand, the inefficiency theory and the
difference theory suggest more positive outlooks. The inefficiency theory suggests the
consequence of a similar but less efficient spatial experience in contrast to that of the sighted
individuals while the difference theory suggests the consequence of a different but functionally
equivalent experience. While evidences for both inefficiency theory and the difference theory
exist in abundance, it is often suggested that the lack of visual experience does not prevent one
from obtaining spatial knowledge because information from each of the different sense is
specialized but also complementary and overlapping or in short, the visually impaired’ s lack in
experience with visual cues can be complemented by obtaining similar information from other
senses though a bias in the system can be induced [16].
It is specifically inferred that as opposed to the deficiency theory, the poor performance by the
congenitally blind compared to those who are sighted are usually due to factors like restricted
capability in spatial processing or storage and the use of relatively less efficient information
processing strategies. Many literature reviews and past studies on the same subject suggest a
correlation in whether an individual is congenitally blind and the strategies implemented
instinctively during navigation. For example, in an experiment involving the exploration of a
largescale layout, the late blind and blindfolded sighted participants, who implemented the back-
andforth method, were able to perform better than the congenitally blind because their strategy
allows the formation of a more integrated representation of a space. Such experiment supports
that the processing strategy selected by the visually impaired - instead of their abilities - is what
makes them less efficient than the sighted in navigation. That is, the visually impaired population
usually choose sequential representation over map-like representation, which could be what
impacted their performance in making spatial reference and caused them to be less efficient that
the sighted individuals. In addition, the visually impaired has the tendency to map cognitively in
reference to their own body over a reference that exists externally. The visually impaired
individual prefers the strategy because it is the most reliable approach in existence for the
obtention of spatial information under blind conditions, but if a task appears to be involved with
mental reorganization, mental rotation or spatial inference, the approach becomes less effective in
contrast to that utilized by the sighted individuals [17-19] .
2.2. Machine Learning: SVM and Regression Algorithms
Machine learning is commonly defined as computational methods that input and analyze existing
data for performing the most efficient yet accurate prediction. It implements sets of machine
learning algorithms to complete tasks by running through a representational process of training,
validating, and testing data samples. The standard learning tasks implemented in this study are
regression and support vector machines (SVMs) [13,20].
a. Support Vector Machines (SVMs)
SVMs are machine learning technique developed to implement the characteristic of kernel for
analyzing samples and classifying new profile into labels. In the use of SVM, a hyperplane is
implemented between classifications in a high-dimensional space to analyze and separate existing
samples. Such plane is selected to locate at a position where the adoption of maximum margin
would be fulfilled so as to maximize accuracy in sample-based future predictions. Two
characteristics- the soft margin and the kernel function- are also implemented within SVM to
ensure the capture of the most data with adoption to minor misclassification. Specifically, the soft
margin introduces a parameter that permits control over the extent of misclassification in
algorithm while the kernel function allows computation with linear method in high-dimensional
space for optimized simplicity. As the most essential in the technique of SVM, the kernel

5
functions perform the major process of adding additional dimensions so as to better classify
samples into labels [14]. While it may result in the overfit of data, kernel makes SVM
competitive when compared to other machine learning techniques due to first, its capability to
project data in higher dimensional space. Second, its capability to construct models of greater
complexity in real-world application and, lastly, its capability to implement optimal simplicity so
to achieve good performance in mathematical analyzation [15].
b. Regression
Regression is a machine learning technique implemented mathematically for the analysis of
relationship between variables and for predicative purpose. Essentially, the technique measures
the extent of effect one variable has on another for building an optimal model that best illustrates
the dependency or independency between variables. Such technique is usually used for the two
major theories of prediction and evaluation of casual relations between variables. The general
definition for regression model is rather broad, as it points to all case where independent variable
predict the dependent without precise direction that specifies on the quantities of variables put in
use or the degree of polynomial implemented. To analyze in detail, the broad characterization of
regression models can in fact be broken into smaller fragments, which would thus produce simple
linear regression, multivariate linear regression (MLR), and polynomial regression. Despite their
share in some common characteristics like the adoption to the least square method, these different
regression models differ distinctively. The simple linear regression is inputted with only a single
independent variable while MLR portray relationship for multiple independent variables. On the
other hand, the polynomial regression differs significantly when compared to both MLR and
simple linear regression in that it analyzes on a relationship modeled by nth degree polynomial
while the former two regressions are modeled linearly [21-22, 26].
2.3. Existing Systems
A. Helal proposed a navigation system called Drishti for the visually impaired population through
the integration of hardware like wearable computers and software built with a web of
technological components such as spatial database and map server. Similar to our application,
Drishti provides a significant amount of guidance to the user through voice support and addresses
their needs through the active use of a digital map with GPS. Drishti readily involves the
navigational features to support the visually impaired population, but our application extends
support to disaster alert and the prevention of COVID-19 [11].
I.Ulrich proposed a navigation device called GuideCane to guide the visually impaired population
during independent navigation. GuideCane is built with a cane on a wheel, and it implements a
steering servo motor with an ultrasonic sensor for the main functionality to operate, which is to
guide the users away from hazards and obstacles through the steering commands. While
GuideCane performs its navigational feature on a wheel, our device operates on its own and is
lighter and can be more portable for the users to bring everywhere for navigation [12].
T.Kiuru proposed a wearable mobility assistive device that implements radar technology. The
operation of such device is based on an electromagnetic radar sensor that features at 24 GHz.
During navigation, radio waves are transmitted and conveyed to the user by audial and/or haptic
means to signal of the direction and distance of possible obstacles in front of the user. The device
also features compensation for inclination errors through its ability to perform automatic
inclination and orientation sensing. While the device is tested to be a potential for its portability,
it is restricted to support in mobility [23].

6
3. CHALLENGES
A few technical challenges have been identified and listed as follows. We will present the
solution in Section 4.
3.1. Challenge 1: Unsafe and Exclusive Urban Design
In today’s society, the layout of cities and the interior design of many buildings are planned and
implemented to closely match with the major population’s aesthetic perception and their desire
toward a sense of freshness. While the aesthetic within these designs brings pleasure to a great
majority, a building’s exterior appearance in the society’s vision has unreasonably dominated
over the many other factors that should’ve been more significant or, to be specific, safety and
convenience. To the many who are blessed with clear vision and have the ability to navigate
freely in spaces, the domination of appearance over safety and convenience in designs does not
seem to be a problem. However, to the visually impaired population who have nothing to rely
upon when navigating in an unfamiliar space and only spatial memory to rely upon in a familiar
space, a simple yet safe interior layout design saves them time and decreases their risk of injury.
The unsafe and inconvenient city design has been the major factor which discourages the visually
impaired population to enjoy the same quality of life and to travel freely around a city, so a
solution has to be proposed to create a more friendly environment for the independent navigation
of the visually impaired population.
3.2. Challenge 2: Lack of Preparedness and Communication in Emergency
Situation
Humans are usually alerted about dangerous presences through their five senses - sight, hearing,
smell, taste, and touch. During unexpected natural disasters, those nearest to the situation observe
the danger to make judgement about what to do next while those who live farther away yet will
possibly be impacted receive warnings on smart devices to prepare ahead of time. The visually
impaired population, however, lack information to make judgments in face of the sudden onset of
natural disasters or to prepare for such situations due to their vision loss, so a solution has to be
proposed to take their community’s safety into consideration during such crises.
3.3. Challenge 3: Sanitization During and After the Pandemic
The emergence of COVID-19, also known as the coronavirus disease, is a disease that leads to
symptoms like dry cough, fever, and tiredness. Though this pandemic has introduced challenges
to everyone, newly enforced rules like social distancing and the adjustments newly made to many
markets have been particularly tough for the visually impaired population to adapt. [8] According
to an article from The Conversation, a public media source in the United Kingdom, the changes
in operation in markets has created difficulties for the visually impaired individuals to navigate
using a spatial map and that social distancing has been difficult for them to employ in some
situation due to their vision loss or inability to notice when people are walking overly close to
them. [9] Moreover, the World Health Organization (WHO) mentions that the visually impaired
population are potentially at a greater risk to contract COVID-19 due to their need to frequently
touch items near them for directions during navigation. To limit the spread of COVID-19 in the
visually impaired community and the whole society for a more immediate end to this pandemic, a
solution must be proposed to guide the visually impaired population about the best method of
navigation during this time and to develop procedures for limiting the chance of being exposed to
COVID-19.

7
4. SOLUTION
4.1. Overview to the Solution
The application has been implemented using MIT App Inventor, and it is carefully developed to
serve as a multi-functional platform to support the visually impaired population in navigations,
during natural disasters, and in the midst of the COVID-19 pandemic. The application intends to
take all aspects into consideration when it provides features like QR code login, locative marker
placement, vibration when detected obstacles, alert in face of disasters, GPS-frequency database,
and sanitization reminder.
Figure 1: Overview of the solution

8
Figure 2: Overview of the solution (Flowchart)
4.2. QR Code Login System
Since the traditional login procedure of typing username and password can provide
inconvenience to the visually impaired users and leads to reduced user experience, our
application intends to prevent such inconvenience by allowing logins through QR codes. After
the user signs up for an account, they can shake the smartphone to activate the QR code login
system, which allows them to directly scan the code to login. The system also comes with voice
guidance, which will provide feedback in response to successful login, allowing communication
to happen directly between the application itself and the visually impaired users.

9
Figure 3: QR Code Login System
In our application, the Global Positioning System (GPS) is employed and used to display the
user’s current location on a map and in text. Longitude, latitude, and a detailed line of address are
also displayed on screen after occurrences of new rounds of syncing. Through Bluetooth
connection, this application can also connect to a smart cane which performs vital tasks like
vibration in detection of obstructions and marker placement. The device -smart cane- itself is
built using a proximity sensor, some wires, and a button from the Arduino kit. Shown in Figure 4,
a proximity sensor is mounted near the tip of the smart cane to detect the surrounding or,
specifically, the path before the user. Whenever the proximity sensor finds an obstruction that can
distract and change the user’s planned path for navigation, it will send a signal to the vibration
motor to vibrate and through it, alert the user to change direction. Shown in Figure 2, the
detection of an obstruction will also increment the frequency value displayed in the application
by one, keeping a record of the frequency or how many times a visually impaired population
would have its device detect an obstruction ahead of steps at different locations. Last but not
least, the device has a button mounted on the very top. Whenever the button is pressed, a tiny red
pin will be placed on the map. This can help the many users on the platform to mark and share
locations that are inconvenient or unpleasant to navigate near at.
Figure 4: Global Positioning System

10
4.3. GPS Coordinate
Our application encourages the users to navigate freely and, indeed, safely. With the global
positioning system, the application accompanies the user on their road and requests consent to
save their GPS coordinates with the device’s corresponding acquired frequency at a specific
location into a database. This will eventually produce a large database that can map out the
locations that are more friendly for the visually impaired population to navigate nearby and, most
importantly, raises the society’s awareness toward making the environment more convenient and
welcoming to every population, which would thus encourage entrepreneurs in the future to learn
from those locations that succeed in making their spaces inclusive and accepting to everyone.
4.4. Natural Disaster Alert System
The frequency, along with the current longitude and latitude of the user, perform vital roles in the
alert system for natural disasters. During a natural disaster, requests will be sent to the users’
device to request for their data on frequency and GPS coordinate. With such information, services
behind the calls will be able to make announcements and assign more personalized procedures to
self-protect based on where the user is at and what resources are accessible near them.
4.5. COVID-19 Sanitization Reminder
Tracking frequency within the application also assists with the function of reminding the user to
sanitize the device during the time of COVID-19 pandemic. Whenever a specific value of
frequency is reached that the device has a high risk of having the virus on its surface, the sanitize
button will light up on the application’s home page while a voice will be played to remind the
user to sanitize the device for minimizing the chance of contracting COVID-19.
4.6. Machine-learning
Dummy data were generated for machine learning in our program for testing and figuring out the
most appropriate machine learning model through the use of svm and regression algorithms for
the following three cases:
 use latitude, longitude, and the three categories of locations (home, store, park) to predict
frequency.
 use longitude, latitude, and frequency to predict between the three categories of locations
(home, store, park).
 use frequency to predict whether a user should sanitize the device.
5. EXPERIMENT
To evaluate the accuracy of the application’s algorithms, we experimented with the different
models, parameters, and data set features to find the most accurate machine learning model. The
first experiment was conducted to find a machine learning model that best describes the
relationship between the user’s GPS Coordinates and the frequency of which an obstruction will
be detected by the user’s device in an one hour walk. This experiment evaluates and compares the
accuracy of the three machine learning models created with 1000 datasets with different
polynomial parameters. The second experiment, otherwise, is conducted with the intake of 1000
datasets with different set features to find a machine learning model that best predicts the user’s
presence in one of the three categories of locations (home, park, store). Lastly, the third

11
experiment also consumes 1000 data sets with different set features for the purpose of predicting
whether the user should sanitize their device at times. In each of the three experiments, the
accuracy of different models is calculated and compared among each other to find the best
representation in each that executes the most accurate prediction.
For experiment 1, machine learning models with different polynomial parameters were applied to
the same data set to find out which model would produce the most accurate algorithm. A linear
model, along with polynomial models with parameter 2, 5, and 7, are compared together to
evaluate the model that produces the most accuracy.
Figure 5: polynomial parameters influence on the model
The results of experiment 1 reveals neither being linear or polynomial nor changing the
polynomial parameters influence the accuracy of the model. Shown in Figure 5, the accuracy of
the linear model, along with that of the polynomial model with parameters of 2, 5, and 7, are all
at a value of 0.87 or a percentage of 87%.
In experiment 2, three machine learning models that differ by their data sets are compared among
each other for determining the one that is most accurate at predicting. The three models take in
datasets of longitude, latitude, and frequency in different combinations, with the first model
intaking three inputs (longitude, latitude, frequency), the second model intaking two inputs
(longitude, latitude), and the third model intaking a single input (frequency).
Figure 6: data sets influence on models Figure 7: different data sets influence on four models

12
The results of experiment 2 reveals that the accuracy for intaking the full data set which contains
longitude, latitude, and frequency is lower than that when intaking a single dataset of frequency,
as the accuracy for model 1 is shown to be at a value of 0.94 or a percentage of 94 % while that
of model 3 is a value of 0.95 or a percentage of 95% . Amongst the three models, however, the
model that intakes two inputs (longitude and latitude) differ the most and produces the least
accuracy to a value of about 0.48 or 48%. [Figure 6]
A total of four machine learning models with different data sets are compared against each other
in experiment 3, in which the first model is created with two inputs (longitude, latitude), the
second model with one single input of the three categories of location (home, park, or store), the
third model with one single input of frequency, and the fourth with a full dataset of four inputs
(longitude, latitude, frequency, categories of location).
The results of experiment 3 reveals that, like what happened in experiment 2, an increasing
amount of intake dataset positively influences the accuracy of its model only in some cases. To
support, Figure 7 displays the model with the most input (Figure 7 Feature Set 4) - longitude,
latitude, frequency, and categories of location - as having the highest accuracy in a decimal of
0.98 and a percentage of 98 %. However, it is also significant to notice that the model producing
the lowest accuracy is the one with intake of two inputs (Figure 7 Feature Set 1) instead of intake
of one single input. To be specific, the model with longitude and latitude as the only inputs
produces an accuracy in decimal of approximately 0.57 and in a percentage of 57%, which is
significantly lower than that produced by the models with one single intake in inputs or, to
emphasize, an accuracy of greater than a value of 0.8 or percentage of 80%. Lastly, the
comparison of accuracy between model 2 and model 3 in Figure 7 emphasizes the difference in
the accuracy of the same amount of input but different input or, in other words, between
predicting based on a single input of the three categories of location from model 2 and based on a
single input of frequency from model 3. To compare, the machine learning model with frequency
as its input has a higher accuracy in decimal of 0.96 or percentage of 96 % than the machine
model with categories of location as its single input, which has a lower accuracy of 0.84 in value
and 84% in percentage. [Figure 7]
Figure 8: Display of all results from experiment
6. RELATED WORK
J. M. Loomis emphasized the system of spatial display, including but not limited to body
pointing, HPI-speech, virtual-tone display, and virtual-speech display, to be an effective system

13
for navigational purpose of the visually-impaired population and proposed that virtual-speech
display, regardless of its blockage toward environmental sounds, performs the best in leading to
the fastest mean travel times and is rated subjectively as the top among the virtual displays used
during the study [10].
Y. Zhang conducted a study to evaluate four regression models – which are standard Poisson
regression, autoregressive adjusted Poisson regression, multiple linear regression, and seasonal
autoregressive integrated moving average (SARIMA) model- and to identify the most appropriate
one for modeling association between climate variation and salmonellosis transmission. The
study was done through the comparison of parameter estimation, goodness-of-fit, and forecasting
ability of the four models based on the notified salmonellosis cases in 2004. The study’s result
reports that though all models demonstrate consistent positive or negative association, the
SARIMA model is the most appropriate to be adopted for the quantification of relationship
between climate variations and salmonellosis transmission for its ability to integrate and control
seasonal variation, autocorrection and long-term trend [24].
F. Laanaya studied and made comparison between a generalized additive model (GAM), logistic
regression, residuals regression and linear regression to identify the one that best demonstrates
the relationship between the dependent variable (average water temperature) and independent
variables (daily average air temperature and average discharge at the Sainte-Marguerite River,
Northern Canada). It is concluded in the end of the study through evaluation over each
implementation’s root mean square error (RMSE), bias error, Nash-Sutcliffe coefficient of
efficiency (NSC), and goodness of fit that GAM performs the best due to its flexibility [25].
7. CONCLUSION AND FUTURE WORK
In conclusion, our application integrates with a Bluetooth-enabled device to encourage the
visually impaired population to navigate independently, alert and guide them in natural disasters,
and remind them to sanitize their devices during the pandemic. In our experiments, we tested
trials of different machine learning models which differ by regression model, polynomial
parameter, and inputted data sets. The first experimentation results show that since adjusting the
regression model or polynomial parameter does not change the accuracy of the prediction at all,
it’s proper to just use a linear regression model to train and test for making predictions on
frequency based on longitude, latitude, and the three categories of locations (home, park, store).
The second and third experimentation results show that the machine learning model that intakes
one single data set of frequency will predict among the three categories of locations the best and
that the machine learning model that takes in the full data set of frequency, longitude, latitude,
and the three categories of location will predict most accurately in whether or not to sanitize the
device. The accuracy of each experiment when using the most appropriate machine learning
model when making predictions all exists above 90% or a value of 0.90, suggesting that the
application will run well and accurately to ensure a pleasant and convenient experience for the
visually impaired population.
While our application exceeds the navigational feature and remains unique from other prototypes
in that it encourages independence in other aspects, it also has some limitations. Since the
application interacts with the user solely through voice, it cannot provide the population that is
both visually impaired and hearing-impaired the same quality of support. In the future, we plan to
improve the accuracy of the system by boosting the signal connections between the device and
API, strengthen the practicality by increasing the device portability, and enhance the optimization
by adding a “help” button. We strongly believe that by following these measures, our alert system
will be able to reach its full potential.

14
REFERENCES
[1] Tracy L Mitzner, PhD, Jon A Sanford, March, Wendy A Rogers, PhD, Closing the CapacityAbility
Gap: Using Technology to Support Aging with Disability, Innovation in Aging, Volume 2, Issue 1,
January 2018, igy008, https://doi.org/10.1093/geroni/igy008
[2] “Disability Impacts All of Us Infographic.” Centers for Disease Control and Prevention, Centers for
Disease Control and Prevention, 16 Sept. 2020,
www.cdc.gov/ncbddd/disabilityandhealth/infographic-disability-impacts-all.html.
[3] “Assistive Technology.” World Health Organization, World Health Organization, who.int/news-
room/fact-sheets/detail/assistive-technology.
[4] Wu, Y., Lindsay, S., Cable, J., Jones, R., Evans, L., &Xie, X. (2018). Digital Media Usage of Sensory
Impaired Users in Wales 2018 Report.
[5] Attia, I., &Asamoah, D. (2020). The white cane. Its effectiveness, challenges and suggestions for
effective use: The case of Akropong School for the Blind. Journal of Education, Society and
Behavioural Science, 47-55.
[6] Whitmarsh, L. (2005). The benefits of guide dog ownership. Visual impairment research, 7(1), 27-42.
[7] Andrew Zaleski, special to CNBC.com. “Amazing Electronic Glasses Help the Legally Blind See, but
They Are Costly.” CNBC, CNBC, 20 Aug. 2018, www.cnbc.com/2017/09/20/theseamazing-
electronic-glasses-help-the-legally-blind-see.html#:~:text=Amazing electronic glasses help the legally
blind see, but they are costly,-Published Wed, Sep&text=Smart glasses, called eSight3, help,price tag
of about $10,000.
[8] Senior Lecturer in Psychology. “The Pandemic Is Undermining Visually Impaired People's
Independence – Here's How to Fix This.” The Conversation, 14 Jan. 2021, theconversation.com/the-
pandemic-is-undermining-visually-impaired-peoplesindependence-heres-how-to-fix-this-142209.
[9] “Coronavirus.” World Health Organization, World Health Organization, www.who.int/health-
topics/coronavirus#tab=tab_1.
[10] Loomis, J. M., Marston, J. R., Golledge, R. G., &Klatzky, R. L. (2005). Personal guidance system for
people with visual impairment: A comparison of spatial displays for route guidance. Journal of visual
impairment & blindness, 99(4), 219-232.
[11] Helal, A., Moore, S. E., & Ramachandran, B. (2001, October). Drishti: An integrated navigation
system for visually impaired and disabled. In Proceedings fifth international symposium on wearable
computers (pp. 149-156). IEEE.
[12] Ulrich, I., &Borenstein, J. (2001). The GuideCane-applying mobile robot technologies to assist the
visually impaired. IEEE Transactions on Systems, Man, and Cybernetics-Part A: Systems and
Humans, 31(2), 131-136.
[13] Mohri, M., Rostamizadeh, A., & Talwalkar, A. (2018). Foundations of machine learning. MIT press.
[14] Noble, W. S. (2006). What is a support vector machine?.Nature biotechnology, 24(12), 15651567.
[15] Hearst, M. (Ed.), Dumais, S.T., Osuna, E. Platt, J. Scholkopf, B., Support Vector Machines, IEEE
Intelligent Systems, Trends & Controversies feature, 13 (4), July-August 1998.
[16] Millar, S. (1994-10-27). Understanding and Representing Space: Theory and Evidence from Studies
with Blind and Sighted Children. : Oxford University Press. Retrieved 11 Jul. 2021
[17] Ungar, S. (2000). “Cognitive mapping without visual experience,” in Cognitive Mapping. Past,
Present, and Future, eds R. Kitchin and S. Freundschuh (London: Routledge), 221–248.
[18] Foulke, E. and Hatlen, P.H. (1992). A collaboration of two technologies. Part 1: Perceptual and
cognitive processes: their implications for visually impaired persons. British Journal of Visual
Impairment, 10, 43-46.
[19] Thinus-Blanc, C. and Gaunet, F. (1997) Representation of space in blind persons: vision as a spatial
sense? Psychological Bulletin, 121, 20-42
[20] Shalev-Shwartz, S., & Ben-David, S. (2014). Understanding Machine Learning: From Theory to
Algorithms. Cambridge: Cambridge University Press.
doi:10.1017/CBO9781107298019
[21] Maulud, D., &Abdulazeez, A. M. (2020). A Review on Linear Regression Comprehensive in
Machine Learning. Journal of Applied Science and Technology Trends, 1(4), 140-147.
[22] Ostertagová, E. (2012). Modelling using polynomial regression. Procedia Engineering, 48, 500-506.
[23] TeroKiuru, MikkoMetso, MikkoUtriainen, KirsimarjaMetsävainio, Hanna-Mari Jauhonen,
RiittaRajala, ReetaSavenius, MirjaStröm, Tiia-Nina Jylhä, Reijo Juntunen&JuhaSylberg | Duncan

15
Shepherd (Reviewing Editor) (2018) Assistive device for orientation and mobility of the visually
impaired based on millimeter wave radar technology—Clinical investigation results, Cogent
Engineering, 5:1, DOI:10.1080/23311916.2018.1450322
[24] Zhang, Y., Bi, P., & Hiller, J. (2008). Climate variations and salmonellosis transmission in Adelaide,
South Australia: a comparison between regression models. International journal of biometeorology,
52(3), 179–187. https://doi.org/10.1007/s00484-007-0109-4
[25] Laanaya, F., St-Hilaire, A., &Gloaguen, E. (2017). Water temperature modelling: comparison
between the generalized additive model, logistic, residuals regression and linear regression models.
Hydrological Sciences Journal, 62(7), 1078-1093.
[26] Lindley, D. (1991). Regression analysis: Theory, methods and applications, by A. Sen and M.
Srivastava Pp 347. DM88. 1990. ISBN 0-387-97211-0 (Springer). The Mathematical
Gazette,75(472), 248-249. doi:10.2307/3620306

AN INTELLIGENT SYSTEM FOR THE ENHANCEMENT OF VISUALLY IMPAIRED NAVIGATION AND DISASTER ASSISTANCE USING GEO-BASED POSITIONING AND MACHINE LEARNING

More Related Content

Similar to AN INTELLIGENT SYSTEM FOR THE ENHANCEMENT OF VISUALLY IMPAIRED NAVIGATION AND DISASTER ASSISTANCE USING GEO-BASED POSITIONING AND MACHINE LEARNING (20)

More from ijgca (20)

Recently uploaded (20)

AN INTELLIGENT SYSTEM FOR THE ENHANCEMENT OF VISUALLY IMPAIRED NAVIGATION AND DISASTER ASSISTANCE USING GEO-BASED POSITIONING AND MACHINE LEARNING