Therefore, it can be concluded that there have
been lots of precise system and programs that are being developed for specific
works but it has not been proven that whether the techniques used in the
developed systems and programs can also be used for other purposes and also
whether it will be as precise for the other tasks as it was for the task for
which it was actually used or developed.
The technique used in developing the program for playing the game of
Go the has proved to be impeccable. But even after its high precision, it
cannot be limpidly claimed that the same technique whether will prove to be as
precise as it has proved for the game. For example, if the same technique is
used for place recognition then will it be as precise as it was for playing the
game of Go.
Another novel system developed for the visually impaired people
specifically for indoor navigation is an innovative system but the system is a
smart one, not an intelligent one. Moreover, for this, a person always needs to
carry the smartphone in which the application is installed. So, that makes it
bit complicated and also bit expensive and also the system demands the visually
impaired person to operate smartphone, so that cannot be considered practical.
The system developed for visually impaired people using the 3600
camera has a precision of 92.8% but it takes around 7 seconds for completing
the whole process which is too long. This certainly cannot be used by the
visually impaired people because it takes a lot of time and before it will
produce an output the person would have moved to a different scenario.
Therefore, the system does nothing to abate any unwanted accident that the
visually impaired person may meet.
However, there are some of the things that are needed to be cleared.
Like when MCDNN is implemented for recognising the traffic signal but it has
not been made pellucid that whether the developed MCDNN can be used for
recognizing other required things like different objects, or not.
In the earlier section, the different research developments have
been portrayed. The precision of above 99% have been achieved and also many of
the latest devices, programming and methods have been used for achieving the
Discussion and Conclusion
Further, a very complex game known for its colossal search space,
the game of ‘Go’. For this game, an intelligent program has been developed with
very high precision such that it can play the game and win it quite easily. The
program developed for this task is named as ‘AlphaGo’. The accuracy rate
achieved in this was 99.8%, it defeated the professional player and champion of
this game in all the games played. There are two basic things in this
development, the first one is the value networks that are being implemented for
analysing the position of the board and the second is policy networks which are
implemented for choosing the which move to make and these have been amalgamated
with Monte Carlo simulation which engendered a new search algorithm. The
learning methods used for making the program intelligent are the supervised
learning and reinforcement learning 9.
A same system for visually impaired people had been developed so
that they can easily navigate in indoor space. In this the IP cameras have been
used that are installed in the home premises. The cameras capture the image and
sends it to the application installed in the smart phone used by the concerned
person. The application using the image processing finds out the objects
present in it and then informs the concerned person by producing an audio
Another development had been for the visually impaired people. The
system has been developed to assist the visually impaired people by providing
them the information about the surroundings. In this system a 3600 camera
is used, that captures the scenario and sends it to the mobile device. To make
the system intelligent the convolutional neural network (CNN) and an already
trained VGG-19 network had been used, so as to recognize the objects in the
captured image. Then the complete message is given to the user (person) in an
audio manner. The accuracy attained for this particular system is 92.8% 7.
Using the discussed concepts, a traffic signal recognition system
had been developed that has an accuracy of 99.46%, which is more than that of
human which is 98.84%. This was achieved by developing the multicolumn deep
neural network (MCDNN), which is nothing but an artificial neural network being
developed using multiple deep neural networks for acquiring robustness. In this
case, it was formed using 5 per pre-processing ways and 25 nets. Also, the
error occurring in recognition task reduced three times. The recognition rate
increased from 98.52% to 99.46% when the system was used with 25 nets. This was
tested practically and won the German traffic sign recognition benchmark 6.
In this section succinct expository the various research and
development that are being done and are riveting.
Therefore, machine learning concept is the backbone for creating an
intelligent system because it actually helps the system in developing a proper
logic so as to acquire a particular way of learning different things and then
Model-based learning is the way in which the system creates models
of different examples or in other words groups, the examples under different
categories and then using these groups or models predicts a suitable output for
the input 5.
Instance-based learning is the way in which the system learns from
some examples and tries to predict an output for an input by associating it
with the examples that it already learned.
Online learning also termed as incremental learning is just the
opposite of what batch learning is. So, a system having the quality of online
learning can take in new data in a particular sequence and can learn from them
even after its production.
Batch learning which is also termed as offline learning, in which a
system is provided data during its production as it cannot take any further
data once its production is complete, in other words, the system which cannot
learn incrementally. If a new data set is needed to be implemented when a new
version of the system is required to be developed.
Reinforcement learning is a way in which a system learns by
observing and analysing the things happening and then doing it by the
experience of observation that it did.
Semi-supervised learning can be defined as an amalgamation of the
supervised and unsupervised learning system.
Apart from these two, there are other types of learning like
semi-supervised learning, reinforcement learning, batch learning, online
learning, instance-based learning, model-based learning and recommender system.
When a system is made to learn different things using a particular
data set then it is supervised learning and when a system learns without the
prior knowledge of what the output will be then it is unsupervised learning.
There are many types of machine learning, however, broadly they are
classified into two categories, supervised learning and unsupervised learning.
Machine learning is the way of developing a learning algorithm that
can be embedded in a system so as to make it an intelligent one. Machine
learning has become very popular as it can be used in the place where no
programming logic can actually help. The same is evident in many of the cases
like spam detection in e-mails, tagging photographs on Facebook, and so on.
Figure 1 shows the schematic of the artificial neural network. From
the figure, it can be depicted that an artificial neural network has an input
layer, an output layer, and hidden layer. There may be many inputs and also
many outputs and also there may be N number of hidden layers. All hidden layers
are connected to each input and also each unit of the hidden layer is connected
to each unit of another hidden layer. The hidden layers can be one to many. The
more the number of hidden layers the more is the complexity and more is the
accuracy, and when the number of hidden layers are more, then they are
specified as deep neural networks. So, the figure completely explains how an
artificial neural network looks like and how are the units (neurons) are
connected to each other. However, it must be noted that the hidden layers are
invisible, i.e. practically they cannot be seen 4.
Fig. 1 Artificial Neural Network
These neural networks are trained for some particular task several
times so as to perform the assigned task precisely. The more the training is
the more impeccable the system will be 3.
The complex networks formed by joining different units (neurons) and
where each unit is dependent on another is termed as an artificial neural
network. The concept has been coined out of the natural neural systems of
humans. Such that it imitates the function of natural neural network and a
system similar to that of humans can be formed.
Artificial Neural Network
So, the concepts present in artificial intelligence are responsible
for the development of intelligent systems or making an existing system
Artificial Intelligence involves neural network, machine learning,
genetic algorithm and so on. The interesting fact about artificial intelligence
is that the developer does not have to develop a detailed program, only an
algorithm is enough.
AI can also be defined as a concept that can be used to develop a
system that does everything like humans and without any support of humans.
Moreover, it can be something that is more precise as compared to human because
of the storage capacity and learning ability 2.
Artificial Intelligence is a method of developing a system that can
act rationally and also like a human. Moreover, a system that can also think
rationally and also like a human.
Artificial Intelligence is a process of developing the qualities of
humans in a particular system by thinking in a way of philosophy and
Before moving into the discussion of different works done in this
sector, an elucidation of various techniques used to make the systems
intelligent will be done. Then the different important and consequential works
will be discussed and at last the pros and cons associated with the sector,
that are needed to be addressed will be discussed.
Using disparate artificial intelligence techniques assistive
technology that can detect and recognize objects, recognize locations, can do
face and expression detection, helping the people remain physically fit and so
on have been developed. Assistive Technology has made its presence felt in
every sector of the world, right from the health-related issues to education 1.
were introduced to the modern technologies then they became
assistive technology. For instance, prosthetics are now something that can
actively move like the real or natural ones, wheelchairs are now operated by
the person using it without much effort, i.e. just by using some switches built
on it. So, the whole concept of developing an assistive system has evolved. Now
the assistive technology is evolving to become an intelligent one (an
intelligent system is a system that performs all the assigned tasks on its own)
because of the unlimited demands of humans, who always remain unsatisfied
because they want something much better than the existing one always.
Therefore, the concepts of artificial intelligence (neural network, machine
learning, etc.) are used to develop the assistive system.
Since the emanation of assistive technology, it has evolved
drastically. Specifically, the assistive word is used for something that
assists humans or living beings in some way or the other. For example,
prosthetics. Earlier there used to be prosthetics that were mechanical in
nature, also the wheelchairs. However, these assistive things
Keywords – Assistive
Technology, Artificial Neural Networks, Artificial Intelligence, Machine
Learning, Intelligent systems
Abstract – Today humans are surrounded by different evolving technologies, and
among them, the most common one is the Assistive Technology. Thus, assistive
technology has become one of the most important and vital phenomena in
everyone’s life. And as the time is passing by the demands for it are
burgeoning as humans want their life to be more facile as they want to be
assisted in every work that they do because they want everything to be done in
quick succession as now none of the people has an ideal time. Therefore, it has
become consequential to study it in more detailed manner. Thus, the paper
demonstrates the different milestones that have been achieved in assistive
technology by using different techniques of making it an intelligent system and
also it elucidates the gap that is there even after such extensive work, which
is needed to be addressed by bridging them. This is done so that it can be
understood where actually the assistive technology is today and in which
direction it is needed to be directed.