GLIMPSE -Journal of Computer Science •Vol. 1(1) , JANUARY-JUNE 2022, pp.

7-10

CompUter VISIon And ItS

ApplICAtIon AreAS
1
Neha Bagawari , 2 Tarushi Khanna
Assistant Professor, Ajay Kumar Garg Engineering College, Ghaziabad, UP, India
1

BTech , CSE, III rd Year Student, Ajay Kumar Garg Engineering College, Ghaziabad, UP, India
1
[email protected], 2 [email protected]

Abstract: For many years, creating machines that pos- plying deep learning models to it, accurate identification and
sessed the characteristics of human intelligence, for exam- classification of objects could be done by machinesincluding
ple, the skill to “see”, and interpret the acts of their sur- the reaction to what they see. It is one of the main technolo-
roundings, was a dream for many. But now, the figments gies that enable the intreratcion of the digital world with the
of yesterday have become the facts of today. [1]. Computer physical world. [2]
vision technology is taking a major part in our lives stead-
ily due to the growth of artificial intelligence and the easy II. HISTORY OF COMPUTER VISION
availability of computational power. The experimentation with Computer-Vision started back in
the 1950s. For the first time, it was used to interpret handwrit-
As the term consists of two components, “Computer” that ten digits and alphabets.
is the manmade electronic machine built for performing
various processes, operations, and calculations, from sets About the same time, neurophysiologists attempted to corre-
of instructions directed by software or hardware, and “Vi- late a response in a cat’s brain by displaying an array of im-
sion”, the act or power of sensing with the eyes to conceive ages to it. They discovered that it responded first to the harder
something, Computer vision can be defined as the visual lines or edges and in the view of science, this meant that im-
perception provided to the machines via which machines age processing begins with a simpler portion of the images
try to understand what they see to achieve a goal. [2] like edges. [2]

In simpler words, Computer Vision, a specialized area of 1. Evolution of Computer-Vision

study of artificial intelligence, deep learning, and machine Before the advent of deep learning,to accomplish computer
learning, that uses specialized methods and makes use of vision task manual coding and lot of extra work was required
general learning algorithms. The goal of computer vision to be done by human operators and developers.For instance,
is to understand the contents of digital images. Basically, for performing simple facial recognition, the following steps
this involves the methods of development that attempt to would be required:
reproduce the capability of human vision.
• Create a database: To store individual images of the ob-
Example of Computer-Vision: Google Translation App, ject that were to be tracked in a particular format.
Automatic Cars, Real-time sports tracking, etc. • Images Annotation: For each image, several points, such
as the width of the nose, measure of distance between
Keywords: Computer Vision, CNN, ANN, DIP both the eyes, and various other features or measure-
ments that help to identify the identity of every person
I. INTRODUCTION should be entered.
Computer vision is a branch of artificial intelligence that imi- • Capture new images: Then, to capture new images, the
tates parts of the complexhuman vision system and enables entire measurement process again should be done again,
the computers and systems to visualize and derive meaning- with the key features of the given images as done before.
ful information from the visual world that includes digital
images, videos, and other inputs. So, it can be stated that AI After this manual work, the application does the comparison
enables computers to think, whereas Computer Vision enables between the measurements of the pre-stored images in the
them to visualize, observe and understand. It trains comput- database to the newly entered images and tells if the image
ers to apply interpretation and understanding to the visual corresponds with any of the profiles tracked. In fact, most of
world. Using videos, digital images from cameras, and ap- the work was being done manually and there was very little

7
GLIMPSE -Journal of Computer Science •Vol. 1(1) , JANUARY-JUNE 2022, pp.
7-10

google photos use this technology to categorize the images.

automation involved. the amount of random sampling error Similarly, this technology could be used to automaticallyiden-
in the results is still very large. tify and segregate the images uploaded by users that are ob-
jectionable according to the social media company guidelines.
Computer vision problems could be solved much easily with
machine learning approach. machine learning helped to re- Object tracking - Used to follow or track an object after it
duce the manual coding to a greater extent. Specific patterns is detected. This is often used with objects and their move-
in images could be detected with programs of much smaller ment detection or in real-time video inputs. For example, Au-
length.later they used mathematical and statistical learning tonomous vehicles detect and classify obstacles while driving
algorithms for eg, support vector machine, logistic regression, such as pedestrians, other vehicles, and their respective mo-
linear regression, decision trees, etc for the same. tions to avoid accidents, collisions and obey traffic laws.[7]
Few problems which always used to be a challenge and could Object detection - Image classification is also used to classify
not be solved by basic software development tools were very the image and then detect its presence in a video or image. A
easily solved by ml (machine learning)tools. For example, few examples include damage detection on an assembly line
back then ml engineers created a software that predicted sur- that requires maintenance.
vival windows of breast cancer better than human experts. But
for building the software a lot of effort and time was needed Content-based image retrieval - Computer vision is used to
from the engineers and breast cancer experts. identify, retrieve, and lookup for the images stored in large
databases, based on their content rather than metadata tags as-
deep learning provided analtogether different approach by sociated with them. This technology can be used for perform-
using neural networks for working on machine learning mod- ing digital asset management and can increase the accuracy of
els, which would be a general-purpose function and capable search and retrieval.[2]
of solving any problem representable through instances or
examples. When a neural network is provided with various IV. HOW DOES COMPUTER VISION WORKS.
labeled examples of a particular data, it will be able to ex- Computer vision is a technology that works by imitation of
tract common patterns between those examples and transform the human brain. Similar to a human brain, computer vision
them into a mathematical equation that eventually helps solve systems also rely on patterns to decode individual objects.[5]
the problem further and classify better. [1]
For example, a facial recognition application that uses deep A Computer vision algorithm interprets an image as a series
learning only requires a developed preconstructed algorithm of pixels, with each pixel having its own set of color values. A
and training the algorithm with the examples of the faces of pixel is basically defined as a unit that consists of a combina-
the persons it intends to detect. If enough examples are pro- tion of 3 additive primary colors ie, RGB(Red Green Blue).
vided, the neural network generally detects the faces without This combination may vary in intensities to represent differ-
the need for any further instructions or measurements. ent colors. Colors are stored inside pixels. Computer Vision
models are then trained with a large amount of data consisting
deep learning is the most popular and very effective method of many many pixels — computers process images, add labels
to perform computer vision problems. In most cases, building to objects on them, and find related patterns in the respective
an advanced deep learning algorithm starts from gathering a objects. For example, if we send a million images of different
large amount of labeled training data and tuning its param- cars as input, the computer will analyze and identify patterns
eters such as the layers of neural networks and training ep- that are similar to all cars, and then at the end of this process,
ochs. Compared to previous approaches of machine learning a model is created that results in accurate detection of whether
algorithms, deep learning is both faster and easier to develop a particular image is a car every time we send them pictures.
and computing resources.[4] [2] [1]
III. APPLICATION OF COMPUTER VISION This task is accomplished by, the two most basic terminolo-
A few examples of tasks established using computer vision gies used, those are (1). CNN (Convolutional Neural Net-
are : work) and (2). Deep Learning.
Image classification - A technique that is used to visualize Deep Learning: Advances in machine learning had a great
an image and can be used to classify it (eg -a cat, an orange, a impact on the destiny of computer vision technology. Deep
person’s face). more precisely speaking, it is capable enough learning, in particular, had a very powerful impact on comput-
to accurately classify the given image. For eg., Facebook uses er vision. It made computer vision algorithms highly efficient
this technology to automatically tag people in given photos,

8
 GLIMPSE -Journal of Computer Science •Vol. 1(1) , JANUARY-JUNE 2022, pp.
7-10
in practical life. The emergence of the convolutional neural
network made computer vision possible for general commer- layers can exist between these two layers. they results the
cial or industrial applications and cemented the technology best of dnn models. Its architects include :
as a worthy investment for companies looking to automate
tasks. [8] Convolutional layer
It is the basic building block of a convolutional neural net-
The machine is taught by the deep neural network training work. It determines the output to the given inputs, accord-
process by using a lot of data sets and countless training cy- ingly. this is achieved through special filters that are used to
cles. It follows the bottom-up approach. During this training extract information from the input pictures, known as the ker-
process, the algorithm automatically extracts the relevant fea- nels. this helps in calculating the dot product between the in-
tures in general, for eg - of ‘cars’. This process produces a put and filter values, which in a result builds a 2-dimensional
model that is applied to previously unseen images to produce activation map for that filter. Cnn then learns the filters that
an accurate classification of the same. are activated only when a particular type of feature at some
spatial position of the input is under observation.
Convolutional Neural Networks -A CNN provides the vision
to a machine learning or deep learning model by breaking or Nonlinearity layer
converting the images into its unitary form ie -labeled and nonlinear functions have a curvature When they are plotted
tagged pixels. Labels are used to conduct mathematical op- and degrees higher than one. the main purpose of this layer is
erations on two functions to produce another, a third function to transform the input signal into the output signal which will
is known as convolutions and make predictions about what be used as an input to the next layer.
is been seen. The algorithm is inspired by the neurons in our pooling layer
brain and is designed to recognize the patterns in complex
input data and often performs the best when patterns are rec- Cnn has layers that take the outputs from a neuron at one
ognized in audio, images, or videos. It basically helps you to layer and add it into another individual neuron in the next
build predictive models using huge databases. Some of the layer. Its main purpose is to scale down the size of the repre-
main applications of CNN consist of sentence classification, sentation to reduce the number of calculations and parameters
face recognition, text recognition, object detection, localiza- in the given model.
tion image characterization, etc. [2]
Fully connected layer
As mentioned, the algorithm takes inspiration from neurons Fully Connected layers are standard deep neural networks,
in our brain by using neural networks. A neural network is a whose objective is to build the predictions arrived from the
group of connected I/O units called neurons which consists activations, to be used for regression or classification. It has
of weights and some biases, that are the features of connec- a similar principle as the conventional multi-layer perceptron
tions between the neurons. Weight is basically a parameter neural system (mlp).
that represents the strength of the connection between units
(neurons) and bias is an indication that tells whether neurons VI. BASIC APPLICATION OF OPEN CV IN
are active or inactive. (Weights and bias) OBJECT DETECTION

Instead of general matrix multiplication, CNN uses a math- Object detection

ematical linear operation called convolution in its layers. It It is a very demanding issue as there are many factors that
contains at least one fully connected layer and a convolutional must be considered, for example, lighting conditions, object’s
layer as in a standard multi-layer neural network.[5] shapes, and colors, occlusion, and many more. A few exam-
Some of the basic works of CNN are - ples that use this technique include - arms detection, wild
• Object detection animal detection, miscellaneous object detection, and human
• Faster r-CNN being detection–
• YOLO- you only look once
• Single-shot detector (SSD) Wild animal detection includes research that is based on Ani-
• Retina net mal detection and is helpful for research that is related to the
locomotive behavior of the intended animal, the applications
V. CNN ARCHITECTURE which are used in real life, the applications which are used in
They are the most prominent and most widely used category everyday life, and to spot the dangerous animal in the public
of neural networks. It contains a multi-layer neural network residentiary zone.
architecture, which contains convolutional layers pursued by
fully connected layers. Few ptional layers,ie Subsampling Human being detection is recognition is a method based on
human activities to recognize various activities via the outer

9
GLIMPSE -Journal of Computer Science •Vol. 1(1) , JANUARY-JUNE 2022, pp.
7-10

[3] https://www.verdict.co.uk/computer-vision-timeline/
sensors, for example, the video sensors to improve the fields [4] https://towardsdatascience.com/everything-you-ever-wanted-
of human health and well-being services, by allowing the au- to-know-about-computer-vision-heres-a-look-why-it-s-so-
tomatic checking activities of the patients. awesome-e8a58dfb641e
[5] Dhillon, Anamika, and Gyanendra K. Verma. “Convolutional
Small arm (gun) detection most criminal or illegal actions oc- neural network: a review of models, methodologies and ap-
cur by utilization of hand held arms such as guns, revolvers, plications to object detection.” Progress in Artificial Intelli-
and knives. A few observations state that these arms are also gence 9.2 (2020): 85-112.
[6] https://www.v7labs.com/blog/computer-vision-applications
utilized for differing illegal actions like assault, thievery, etc.
[7] https://viso.ai/computer-vision/the-most-popular-computer-
Automatic strategies for weapon identification are used lately vision-tools/
to avoid and reduce criminal acts. [8] https://www.dynam.ai/what-is-computer-vision-technology/
[9] https://www.analyticsvidhya.com/blog/2021/06/everything-
happening-in-computer-vision-that-you-should-know/

About The Authors

Ms. Neha Bagwari has completed her

graduation from Government College of
Engineering, Amravati, Maharashtra. Post-
graduation she has worked with Cognizant
Technology Solutions and HSBC GLT In-
dia. Combined she has work experience
in Manufacturing and Logistics, Life Sci-
ences, and Banking Domain. Her work ma-
jorly was around Java, J2EE technologies.
She did her MTech from ABES Engineering College, Ghaziabad
affiliated with Dr. APJAK Technical University, Lucknow. Her re-
search subject was “Big Data”. She has experience working with
Fig 1: Machine learning and deep learning workflow the Noida Institute of Engineering and Technology in the education
field. She has published one research article in International Confer-
VII. CONCLUSION ences. She is an Assistant Professor in the Department of Computer
Thus, we have seen that with the advent of deep learning tech- Science and Engineering of AKGEC Ghaziabad affiliated with Dr.
APJAK Technical University, Lucknow. Her present research inter-
niques we have a come long way in computer vision, and have
ests include investigating the different application areas of Big Data
huge application areas where these deep learning techniques and the technologies providing solution to it.
can be used for the purpose of image classification, object lo-
calization, object detection, object tracking, performing trans- Ms. Tarushi Khanna is a student in B.Tech
formation analysis, etc. Third year, CSE Department, AKGEC. Her
interest areas are Data Analytics, Machine
References Learning, Deep Learning, and Artificial
[1] https://xd.adobe.com/ideas/principles/emerging-technology/ Intelligence. She is looking forward to an
what-is-computer-vision-how-does-it-work/ opportunity to start her career as a Data
[2] h t t p s : / / w w w. i b m . c o m / i n - e n / t o p i c s / c o m p u t e r - Scientist and Machine Learning Engineer
vision#:~:text=Computer%20vision%20is%20a%20 along with Computer Vision.
field,recommendations%20based%20on%20that%20informa-
tion.

10