Volume 7, Issue 8, August – 2022 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Analytical Study on Object Detection using

Yolo Algorithm
Dawn Wilson 1, Dr. Manusankar C.2, Dr. Prathibha P H3
PG Scholar1, Assistant Professor and Head2, Assistant Professor 3
PG Department of Computer Science Sree Sankara Vidyapeetom college
Valayanchirangara, Perumbavoor Ernakulam, Kerala

Abstract:- Object detection is a technique that allows vector of probability scores, each representing the probability
detecting and locating objects in videos and images. of the observation occurring at a particular depth.
Object detection is widely used to count objects in a
scene, track their precise locations and accurately label
the objects. It seeks to answer what is the object? and
Where is it? . Object detection adopts various
approaches such as fast R-CNN, Retina-Net, Single Shot
MultiBox Detector (SSD) and YOLO. Among these,
YOLO is the most powerful algorithm for object
detection and as well as suited for real-time scenarios. It
is popular because of its accuracy and speed. YOLO
uses Neural networks to provide object detection.

Keywords:- YOLO, Neural Networks, CNN, Object

detection.

I. INTRODUCTION

Object detection is one of the interesting computer

vision techniques. One of the major application of object Fig. 2.1 Conventional Neural Network
detection is self-driving cars where systems use computer
vision, technologies for the multidimensional III. WORKING OF YOLO
representation of roads and LADAR. YOLO stands for
‘You Only Look Once’. This is an algorithm that detects YOLO includes three techniques. They are Residual
various objects in a picture (in real-time). Object detection Blocks, Bounding Box Regression and Intersection Over
in YOLO is a regression problem and use the class Union[IOU].
probabilities of the detected images.
A. Residual blocks
YOLO algorithm uses convolutional neural networks This is the first step. Here, the image fig:-3.1 is divided
(CNN) to detect objects in real-time. The algorithm into some grids. Each grid has a S * S dimension.
requires only a single forward propagation through a neural
network to detect objects. It is why it is named as ‘You
Only Look Once’. On a single run of the algorithm, the
prediction of entire image is done. CNN is mainly used to
predict bounding boxes and each class probabilities.
Subsequent versions of YOLO include YOLO V1, YOLO
V2, and YOLO V3 up to YOLO V5.

II. CONVOLUTIONAL NEURAL NETWORK

A Convolutional Neural Network (ConvNet/CNN) is

a deep learning algorithm that can assign importance
(learnable weights and biases) to various aspects/objects in
an image and be able to differentiate one from the other.
The pre-processing required in a ConvNet is much less
complicated than in other classification algorithms.

In fig 2.1 Convolutional Neural Networks are a Fig. 1: Grid Dimensions

different type of neural network. The layers are arranged in
three dimensions: width, height, and depth. The neurons in
one layer don't connect to all the neurons in the next layer,
but only to a small region of it. The final output will be a

IJISRT22AUG337 www.ijisrt.com 587

 Volume 7, Issue 8, August – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
A bounding box is a box like outline to highlight The intersection over union ensures that the predicted
objects. It consists of the following attributes: bounding boxes match the real boxes of the objects. This
 Width (bw) process eliminates unnecessary object bounding boxes that
 Height (bh) don't match the characteristics of the objects. The final
 Class (example: car, person, etc.)- This is represented detection will be made by fitting objects perfectly into
by the letter c. predefined bounding boxes.
 Here fig:-3.2 Bounding box center (bx,by)
For example, The car is enclosed by the pink bounding
box while the bicycle is enclosed by the yellow bounding box.
The blue box has been used to highlight the dog.

V. LITERATURE SURVEY

The paper [1] The YOLO algorithm is used to detect and

classify objects in images. It was trained on the classes from
the COCO dataset. YOLO uses a single convolutional network
to predict the likely boundaries of objects in an image Fig:-5.1.
The bounding boxes are weighted according to probabilities,
and the model uses the final weights to determine when to
Fig. 2: Bounding Box detect them. The bounding box represents four dimensions,
including the center of the bounding box, width, height and 4.
B. Intersection over union (IOU) The value of 'c' refers to an object class.Flow chart of YOLO
Intersection over union (IOU) describes how boxes algorithm:
overlap. In fig. 2 it provide an output box that surrounds the
objects perfectly.

Fig. 3: IOU Box overlap

There are two bounding boxes in the above image,

blue and green. The blue box is the predicted box while the
green box is the real box. YOLO ensures that the two
bounding boxes are equal. Fig. 5: Detect and classify objects
IV. COMBINATION OF THE THREE TECHNIQUES Applications of YOLO include Vehicle detection, Crowd
Detection, Optical Character Detection and Image Fire
Detection.

The paper [2] proposes a unified object detection

algorithm using YOLO. They combined separate components
of object detection into a single neural network. Fast YOLO is
the fastest object detector currently available, and it is also one
of the most advanced real-time object detectors out there.
YOLO also generalizes well to new domains, making it ideal
for applications that rely on fast, robust object detection. An
experimental comparison with different object detection
systems has shown that YOLO can be generalized to new
domains better than other detectors.

Fig. 4: single convolutional neural network The paper [3] this new network model suggests a
modified version of the YOLOv1 network model that better
For example 4, There are at least three types of predicts user behavior. The YOLOv1 network is improved by
objects: a car, a dog, and a bicycle. All of the predictions optimizing its 'e loss function. The inception model structure is
are made simultaneously using a single convolutional added. A spatial pyramid pooling layer is used to pool data.
neural network. The proposed model is effective at extracting features from
images, outperforming other models in object detection. It is

IJISRT22AUG337 www.ijisrt.com 588

Volume 7, Issue 8, August – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
trained using an end-to-end method, and has shown REFERENCES
effective results on a challenging Pascal VOC dataset.
[1.] Dr. S.V. Viraktamath, Rachita Byahatti, Madhuri
The paper [4] "YOLOv3: An Incremental Yavagal “Object Detection and Classification using
Improvement" by Joseph Redmon and Ali Farhadi YOLOv3” Dept. of Electronics and Communication
discusses how YOLOv3 is an incremental improvement Engineering SDM College Of Engineering and
over earlier versions of the algorithm. In this paper, the Technology Dharwad, India. International Journal of
authors present some updates to the YOLO algorithm. They Engineering Research & Technology (IJERT) ISSN:
made a number of small design changes to improve it. We 2278-0181 IJERTV10IS020078 ,Vol. 10 Issue 02,
also trained a new network that’s really great.It's a little February-2021
bigger than last time, but more accurate. [2.] Joseph Redmon, Santosh Divvala, Ross Girshick, Ali
Farhadi “You Only Look Once: Unified, Real-Time
The paper [5] "Comparative Research of YOLO Object Detection”, University of Washington , Allen
Architecture Models in Book Detection" by Maria Kalinina Institute for AI , Facebook AI Research.
and Pavel Nikolaev discusses the different YOLO [3.] Tanvir Ahmad , Yinglong Ma , Muhammad Yahya, Belal
architecture models and their effectiveness in detecting Ahmad, Shah Nazir , and Amin ul Haq. Hindawi
books. This white paper describes the creation of a deep Scientific Programming Volume 2020, Article ID
convolutional neural network based on the YOLO 8403262, 10 pages
architecture for real-time book detection. https://doi.org/10.1155/2020/8403262.
[4.] Joseph Redmon, Ali Farhadi, University of Washington,
The Paper [7] “Real-Time Objects” by Priya Kumari, “YOLOv3: An Incremental Improvement”
Sonali Mitra, Suparna Biswas, Sunipa Roy, Sayan Roy [5.] Maria Kalinina, Pavel Nikolaev “Comparative Research
Chowdhury, Antara Ghosal, Palashree Dhar and Anurima of YOLO Architecture Models in Book Detection”
Majumdar. The main goal of this paper is to find the [6.] Priya Kumari1, Sonali Mitra, Suparna Biswas, Sunipa
location of an object in a given picture accurately and mark Roy, Sayan Roy, Chaudhuri, Antara Ghosal, Palasri
the object with the appropriate category. We used a real- Dhar, Anurima Majumder, “YOLO Algorithm Based
time object detection algorithm, YOLO, to train our Real-Time Object Detection”, Volume 8 Issue 1
machine learning model.YOLO is a powerful neural [7.] Zhi Xu , Jingzhao Li , Yifan Meng and Xiaoming Zhang
network that can detect objects in real time and with the “CAP-YOLO: Channel Attention Based Pruning YOLO
help of the COCO dataset. This technique can detect an for Coal Mine Real-Time Intelligent Monitoring”
object in real time with 90% accuracy. [8.] https://towardsdatascience.com/yolo-you-only-look-
The paper [8] CAP-YOLO: Attention-based YOLO once-3dbdbb608ec4
channel reduction for real-time Intelligent Monitoring. In [9.] https://appsilon.com/object-detection-yolo-algorithm/
this paper, CAP-YOLO (Channel Attention based Pruning
YOLO) and AEPSM (adaptive image enhancement
parameter selection module) are proposed to help achieve
real-time intelligent analysis for coal mine surveillance
videos.

VI. CONCLUSION AND FUTURE WORK

Object detection is a Computer Vision and Image

Processing technique that deals with detecting instances of
various classes of objects from the captured image or video.
As compared to other object detection algorithms and
approaches, YOLO is the fast and accurate detection
algorithm. It follows a regression approach, that is, rather
than selecting a field of interest for an image, they estimate
groups and bounding boxes for the whole picture in one run
of the algorithm [1].

Improvements are still making on YOLO. To obtain

better results, the loss function of YOLO can be modified.
Pooling layer can be added. The future and scope of YOLO
object detection is vast

IJISRT22AUG337 www.ijisrt.com 589