Object Single Frame Using YOLO Model

© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 2834
Object Single Frame Using YOLO Model
Mrs. NVN. SOWJANYA1, NENAVATH PARAMESH2, ANUMULA PRANAY KUMAR3 ,
MAAKAM ROSHINI⁴
1Assistant Professor in Department of CSE, Teegala Krishna Reddy Engineering College,
JNTUH University, Telangana, India,
2,3,4 UG Scholar in Department of CSE, Teegala Krishna Reddy Engineering College,
JNTUH University, Telangana, India.
------------------------------------------------------------------------***------------------------------------------------------------------------
Abstract: Our project on this area has been making great
progress in many directions. The main goal of the project is
to detect multiple object in a single frame. In this we have
increased the classification accuracy of detecting objects and
an overview of past research on object detection. You Only
Look Once (YOLO) algorithm is the fastest, most efficient
algorithm and uniquely detects.. In this comparative analysis,
using the Microsoft COCO(Common Object in Context)
dataset, the performance of the algorithm is evaluated and
the strengths and limitations are analyzed based on
parameters such as accuracy and precision.
Keywords: Object Detection, YOLO Algorithm,
Prediction
1. Introduction
A computer views all kinds of visual media as an
array of numerical values. As a consequence of this
approach, they require image processing algorithms to
inspect contents of images. Object detection is a key ability
required by most computer and robot vision systems. Our
project on this area has been making great progress in
many directions. In our project, we have increased the
classification accuracy of detecting objects and we give an
overview of past research on object detection, outline the
current main research directions, and discuss open
problems and possible future directions. You Only Look
Once (YOLO) algorithm correlates activities and uniquely
detects. The fastest and most efficient of algorithm. In this
comparative analysis, using the Microsoft COCO (Common
Object in Context) dataset, the performance of the
algorithm is evaluated and the strengths and limitations
are analyzed based on parameters such as accuracy and
precision. The comparison between Single Shot Detection
(SSD), Faster Region based Convolutional Neural Networks
(Faster R-CNN) and You Only Look Once (YOLO), From the
results of the analysis, YOLO processes images at 30 FPS
and has a mAP of 57.9% on COCO test-dev. In an identical
testing environment, YOLO outperforms SSD and Faster R-
CNN, making it the best of these algorithms. Finally, we
propose a method to jointly train on object detection and
classification. Using this method, we train YOLO
simultaneously on the COCO detection dataset and the
ImageNet classification dataset.
2. Literature Survey
In the recent few years, diverse research work
happened to develop a practical approach to accelerate the
development of deep learning methods. Numerous
developments accomplished excellent results and followed
by continuous reformations in deep learning procedures.
Object localization is the identification of all the visuals in a
photograph, incorporating the precise location of those
visuals. By using deep learning techniques for object
identification and localization, computer vision has reached
a new zenith. Due to significant inconstancies in
viewpoints, postures, dimensions, and lighting positions, it
is challenging to succeed in the identification of objects
perfectly. Accordingly, considerable concern has been given
by researchers to this area in the past few years. There are
two types of object detection algorithms. Object detection
algorithms using region proposal includes RCNN , Fast
RCNN, and Faster RCNN, etc. These techniques create
region proposal networks (RPN), and then the region
proposals are divided into categories afterward. On the
other side, object detection algorithms using regression
includes SSD and YOLO, etc. These methods also generate
region proposal networks (RPN) but divide these region
proposals into categories at the moment of generation. All
of the procedures mentioned above have significant
accomplishments in object localization and recognition.
YOLO consolidates labels in diverse datasets to form a tree-
like arrangement, but the merged labels are not
reciprocally exclusive. YOLO9000 enhances YOLO to
recognize targets above 9000 categories employing
hierarchical arrangement. Whereas YOLOv3 uses
multilabel classification, it replaces the approach of
estimating the cost function and further exhibits
meaningful improvement in distinguishing small targets.
The arrangement of this paper is as follows. Below in
section 2, background information of object detection
methods is covered. It includes two stage detectors with
their methodologies and drawbacks. Section 3 elaborates
one stage detectors and the improved version YOLO v3-
Tiny. Section 4 describes implementation results and
comparison of object detection methods based on speed
and accuracy. Finally, section 5 summarizes the conclusion.
International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 06 | Jun 2022 www.irjet.net p-ISSN: 2395-0072

2.1 Related work:
In this section, we present background
information. It elaborates the most representative and
pioneering two-stage object detection methods with their
significant contributions in object detection. First, we
examine their methodologies and then explain their
drawbacks.
2.1.1. HOG
HOG is a feature descriptor that is extensively applied in
various domains to distinguish objects by identifying their
shapes and structures. Local object structure, pattern,
aspect, and representation can usually be characterized by
the arrangement of gradients of local intensity or the ways
of edges. In the HOG detection method, the first step is to
break the source image into blocks and then distribute each
block in small regions. These regions are called cells.
Commonly, the blocks of image overlap each other, due to
this corresponding cell may be a part of many blocks. For
each pixel inside the cell, it calculates the gradients
vertically and horizontally.
2.1.2 RCNN
Region based convolutional neural networks (RCNN)
algorithm uses a group of boxes for the picture and then
analyses in each box if either of the boxes holds a target. It
employs the method of selective search to pick those
sections from the picture. In an object, the four regions are
used. These are varying scales, colours, textures, and
enclosure.
Drawbacks of RCNN method- Based on a selective
search, 2,000 sections are excerpted per image. For every
region or part of the image, we have to select features using
CNN. For this, if we have 'i' number of images, then selected
regions will become i×2,000. The whole method of target
identification through RCNN utilizes the following three
models: Linear SVM classifier for the identification of
objects, CNN is employed for characteristic extraction, and
a regression model is required to tighten the bounding
boxes. All these three processes combine to take a
considerable amount of time. It increases the running time
of RCNN method. Therefore, RCNN needs almost 40 to 50
seconds to predict the result for several new images .
2.1. FAST RCNN
In place of using three different models of RCNN, Fast
RCNN employs one model to excerpt characteristics from
the different regions. Then it distributes the regions into
several categories based on excerpted features, and the
boundary boxes of recognized divisions return together.
Fast RCNN uses the method of spatial pyramid pooling
to calculate only one CNN representation for the whole
image.
It passes one region for each picture to a particular
convolutional network model by replacing three distinct
models for excerption of characteristics, distributing into
divisions, and producing bounding boxes.
Drawbacks of Fast RCNN method- Fast RCNN also
employ a selective search method to detect concerned
regions. This method is prolonged and demands a lot of
time. Usually, for the detection of objects, this complete
procedure needs almost two seconds for each picture.
Therefore its speed is quite good in contrast to RCNN.
However, if we contemplate extensive real-life datasets,
then the execution of fast RCNN approach is still lacked in
speed.
2.2. Faster RCNN
Faster RCNN is a transformed variant of fast
RCNN. The significant difference between both is that faster
RCNN implements region proposal network (RPN), but fast
RCNN utilizes a selective search technique for producing
concerned regions. In input, RPN accepts feature maps of
picture and produces a collection of object
recommendations and an objectness score per
recommendation in output. Usually, this approach takes ten
times less time in contrast to fast RCNN approach because
of RPN.
Drawbacks of faster RCNN method- To excer2.pt all the
targets in a given picture, this procedure needs multiple
passes for that particular picture. Different systems are
working in a sequence therefore, the performance of the
upcoming operation is based on the performance of
preceding operations. This approach uses region proposal
networks to localize and identify the objects in a picture.

2.3. Table
3. Existing System
Object detection consists of various approaches such
as fast R-CNN, Retina-Net, and Single-Shot Multi Box
Detector (SSD). Although these approaches have solved
the challenges of data limitation and modeling in object
detection, they are not able to detect objects in a single
algorithm run. We have different object detection
algorithms, they are applied based on their accuracy,
precision.
4. Proposed System
It will detect the objects, by using the application of
Linear Support Vector Machine. We apply a single neural
network to the full image. This network divides the image
into regions and predicts bounding boxes and probabilities
for each region. These bounding boxes are weighted by the
predicted probabilities.
YOLO is extremely fast and accurate. In mAP measured at
0.5, IOU YOLO is on par with Focal Loss but about 4x faster.
Moreover, we can easily tradeoff between speed and
accuracy simply by changing the size of the model, no
retraining required.
5. Architecture
Architecture describes how the application is going to
function. Project Architecture of Object Detection describes
how the user’s request is taken as input and how the
output is delivered. The detailed architecture is shown in
Fig 1- Project Architectureof Object detection
Fig 1. Project Architecture of Object Detection
You Only Look Once: Unified, Real-Time Object
Detection. In this it is introduced a new approach to object
detection. The feature extraction and object localization
were unified into a single monolithic block. Further more
the localization and classification heads were also united.
Their single-stage architecture, named YOLO (You Only
Look Once) results in a very fast inference time. The frame
rate for 448x448 pixel images was 45 fps (0.022 s per
image) on a Titan X GPU while achieving state-of-the-art
mAP (mean average precision). Smaller and slightly less
accurate versions of the network reached 150 fps. This new
approach, together with other detectors built on light-
weight Google’s MobileNet backbone, brought the vision
(pun intended) of detection networks and other CV tasks
on edge devices ever closer to reality.
6. Implementation
For realistic execution we are using Operating System
Multi Platform (Windows 7 & above, Linux GCC),and
Backend as Python 3.6 & above. Dataset as COCO (Common
Objects In Context)and Machine Learning Model YOLO V3
(You Only Look Once). Using this application, we can detect
the objects and specify them. In order to use the application
the user has to run the application and can upload a video
file or image file to the program by giving the file path. Itis
designed to detect many objects with the specification. It
can easily find out common objects such as chairs, remotes,
bus etc. The application gives a glimpse, of where the
object is located with the accuracy. The core features of
this project are it provides feedback in the form of video
file or image file, it detects most of the common objects in
context.The other features include detection of each image
is reported with some form of pose information. For
example, for face detection in a face detector system
compute the locations of the eyes, nose and mouth, in
addition to the bounding box of the face.
6.1. Python 3
Type Characteristics time Drawbacks
RCNN To generate regions,
it uses selective
search. From each
picture, it extracts
around 2000
regions.
40-50
sec
Time taken for prediction
is large because several
regions pass through CNN
definitely, and it employs
three distinct models for
the detection of targets
Fast
RCNN
To excerpt the
features, each
picture passes one
time through CNN.
All distinct models
applied in RCNN are
combined
collectively to form
a single model.
2 sec The method used is
prolonged and time
consuming Therefore,
computation time is still
high.
Faster
RCNN
The previous
approach is
replaced with the
region proposal
networks.
Therefore, this
procedure works
much faster
compared to
previous methods.
0.2
sec
Object region proposal is
timeconsuming. Different
types of systems are
operating in sequence.
Thus, the performance of
entire procedure is based
on the working of the
preceding operations.

If you are on Ubuntu, it’s most likely that Python 3 is
already installed. Run python3 in terminal to check
whether its installed. If its not installed use
sudo apt-get install python3
For macOS please refer my earlier post on deep learning
setup for macOS.
I highly recommend using Python virtual environment.
Have a look at my earlier post if you need a starting point.
6.2. Libraries
6.2.1. Numpy
pip install numpy
This should install numpy. Make sure pip is linked to
Python 3.x ( pip -V will show this info)
If needed use pip3. Use sudo apt-get install python3-pip to
get pip3 if not already installed.
6.2.2 OpenCV Python
OpenCV-Python
You need to compile OpenCV from source from the master
branch on github to get the Python bindings.
(recommended)
Adrian Rosebrock has written a good blog post on
PyImageSearch on this.
(Download the source from master branch instead of from
archive)
If you are overwhelmed by the instructions to get OpenCV
Python bindings from source, you can get the unofficial
Python package using
pip install opencv-python
This is not maintained officially by OpenCV.org. It’s a
community maintained one. Thanks to the efforts of Olli-
Pekka Heinisuo.
6.2.3. OpenCV dnn module
DNN (Deep Neural Network) module was initially part of
opencv_contrib repo. It has been moved to the master
branch of opencv repo last year, giving users the ability to
run inference on pre-trained deep learning models within
OpenCV itself.
(One thing to note here is, dnn module is not meant be used
for training. It’s just for running inference on
images/videos.)
Initially only Caffe and Torch models were supported. Over
the period support for different frameworks/libraries like
TensorFlow is being added.
Support for YOLO/DarkNet has been added recently. We
are going to use the OpenCV dnn module with a pre-trained
YOLO model for detecting common objects.
7. Application of YOLO
YOLO algorithm can be applied in the following fields:
7.1. Autonomous driving:
YOLO algorithm can be used in autonomous cars to
detect objects around cars such as vehicles, people, and
parking signals. Object detection in autonomous cars is
done to avoid collision since no human driver is controlling
the car.
7.2. Wildlife:
This algorithm is used to detect various types of animals
in forests. This type of detection is used by wildlife rangers
and journalists to identify animals in videos (both recorded
and real-time) and images. Some of the animals that can be
detected include giraffes, elephants, and bears.
7.3. Security:
YOLO can also be used in security systems to enforce
security in an area. Let’s assume that people have been
restricted from passing through a certain area for security
reasons. If someone passes through the restricted area, the
YOLO algorithm will detect him/her, which will require the
security personnel to take further action.
8. Output Screens
Fig 2. Object detection

Fig 3. Motion Object Detection
9. Conclusion
Here is our project that address problems with existing
system and solves them effectively. In the end, we have
achieved a fully functional RCNN model that efficiently
extracts ships from high resolution satellites images.
Approach helps in increasing the accuracy and speedand
achieves the desired results. By using method, weare able
to detect object more precisely and identify the objects
individually with exact location of an object in the picture
in x, y axis. implementations of the YOLO algorithm on the
web using Darknet is one open-source neural network
frame work.Dark net was written in the C Language, which
makes it really fast and provides for making computations
on a GPU,essential for real-time predictions.
10. Future Enhancement
In future we can add a faster model that runs on the GPU
and use a camera that provides a 360 field of view and
allows analysis completely around the person. We can also
include a Global Positioning System and allow the person to
detect the objects instantly without any delay in frames and
seconds.
11. References
[1] G. Dahl, D. Yu, L. Deng, and A. Acero, “Contextdependent
Pre-trained Deep Neural Networks for Large Vocabulary
Speech Recognition,” IEEE Transactions on Audio Speech &
Language Processing, vol. 20, Jan. 2012, pp. 30-42, doi:
10.1109/TASL.2011.2134090.
[2] D. C. Ciresan, U. Meier, L. M. Gambardella, and J.
Schmidhuber, “Deep Big Simple Neural Nets Excel on
Handwritten Digit Recognition,” CoRR, vol. 22, Nov. 2010,
pp. 3207-3220, doi: 10.1162/NECO_a_00052.
[3] R. Collobert and J.Weston, “A Unified Architecture for
Natural Language Processing: Deep Neural Networks with
Multitask Learning,” International Conference on Machine
learning (ICML 08), ACM press, Jul. 2008, pp. 160-167, doi:
10.1145/1390156.1390177. 12, 05012 (2017) DOI:
10.1051/ 7120 ITA 2017 ITM Web of Conferences
itmconf/201 5012 5
[4] R. Raina, A. Madhavan, and A. Y. Ng, “Large-scale Deep
Unsupervised Learning Using Graphics Processors,”
International Conference on Machine Learning (ICML 09),
ACM press, Jun. 2009, pp. 873- 880, doi:
10.1145/1553374.1553486.
[5] Z. Y. Han and J. S. Chong, “A Review of Ship Detection
Algorithms in Polarimetric SAR Images,” International
Conference on Signal Processing (ICSP 04), IEEE press, vol.
3, Sept. 2004, pp. 2155-2158, doi:
10.1109/ICOSP.2004.1442203.
[6] R. Raina, A. Madhavan, and A. Y. Ng, “Large-scale Deep
Unsupervised Learning Using Graphics Processors,”
International Conference on Machine Learning (ICML 09),
ACM press, Jun. 2009, pp. 873- 880, doi:
10.1145/1553374.1553486.
[7] Z. Y. Han and J. S. Chong, “A Review of Ship Detection
Algorithms in Polarimetric SAR Images,” International
Conference on Signal Processing (ICSP 04), IEEE press, vol.
3, Sept. 2004, pp. 2155-2158, doi:
10.1109/ICOSP.2004.1442203.
[8] K. Eldhuset, “An Automatic Ship and Ship Wake
Detection System for Spaceborne SAR Images in Coastal
Regions,” IEEE Transaction on Geoscience and Remote
Sensing, vol. 34, Jul. 1996, pp. 1010-1019, doi:
10.1109/36.508418.
[9] H. Greidanus, P. Clayton, N. Suzuki, and P. Vachon,
“Benchmarking Operational SAR Ship Detection,”
International Geoscience and Remote Sensing Symposium
(IGARSS 04), IEEE press, vol. 6, Dec. 2004, pp. 4215-4218,
doi: 10.1109/IGARSS.2004.1370065.
[10] C. C. Wackerman, K. S. Friedman, and X. Li, “Automatic
Detection of Ships in RADARSAT-1 SAR Imagery,” Canadian
Journal of Remote Sensing, vol. 27, Jul. 2014, pp. 568-577,
doi: 10.1080/07038992.2001.10854896.
[11] D. J. Crisp, “The State of the Art in Ship Detection in
Synthetic Aperture Radar Imagery,” Organic Letters, vol.
35, May 2004, pp. 2165-2168.
[112] C. Zhu, H. Zhou, R. Wang and J. Guo, “A Novel
Hierarchical Method of Ship Detection from Spaceborne
Optical Image Based on Shape and Texture Features,” IEEE
Transactions on Geoscience and Remote Sensing, vol. 48,

Sept. 2010, pp. 3446-3456, doi: 10.1109/
TGRS.2010.2046330.
Author Profile
<Authors >
Author 1 NVN. SOWJANYA
Author 2 NENAVATH PARAMESH
Author 3 ANUMULA PRANAY KUMAR
Author 4 MAAKAM ROSHINI

Object Single Frame Using YOLO Model

Recommended

More Related Content

Similar to Object Single Frame Using YOLO Model (20)

More from IRJET Journal (20)

Recently uploaded (20)

Object Single Frame Using YOLO Model