Object Detection Using Machine Learning Techniques

A Seminar Report
Submitted to the SPPU, Pune
in partial fulfillment of requirements for the award of degree

Master of Engineering Technology

in
Computer Engineering
by
SOLAVANDE MADHURA TUKARAM
PRN- 77300778E

DEPARTMENT OF COMPUTER ENGINEERING/IT/AIDS

DATTAKALA GROUP OF INSTITUTIONS
SWAMI CHINCHOLI, Tal. Daund, Dist. Pune 413130,
MAHARASHTRA
April 2024
 DEPARTMENT OF COMPUTER ENGINEERING/INFORMATION
TECHNOLOGY/AIDS
DATTAKALA’S GROUP OF INSTITUTIONS SWAMI CHINCHOLI, Tal.
Daund, Dist. Pune 413130, MAHARASHTRA
2023-24

CERTIFICATE

This is to certify that the report entitled Object Detection Using Machine
Learning Techniques submitted by SOLAVANDE MADHURA TUKARAM (
PRN- 77300778E), to Department of Computer Engineeringin partial fulfillment of
the M.E. in Computer Engineering is a bonafide record of the seminar work carried
out by him under our guidance and supervision. This report in any form has not been
submitted to any other University or Institute for any purpose.

Dr. S. S. Bere Dr. D. B. Hanchate

(Seminar Guide) (Head of the Department)

Associate Professor Professor
Dept. of Comp/IT/AIDS Dept. of Comp/IT/AIDS
DATTAKALA’S GROUP OF DATTAKALA’S GROUP OF
INSTITUTIONS, SWAMI CHINCHOLI, INSTITUTIONS, SWAMI CHINCHOLI,
Tal. Daund, Dist.Pune 413130, Tal.Daund, Dist.Pune 413130,
MAHARASHTRA MAHARASHTRA

Principal
DATTAKALA’S GROUP OF INSTITUTIONS, SWAMI CHINCHOLI, Tal. Daund,
Dist. Pune 413130, MAHARASHTRA
 DECLARATION

I SOLAVANDE MADHURA TUKARAM hereby declare that the seminar report

Object Detection Using Machine Learning Techniques, submitted for partial
fulfillment of the requirements for the award of degree of Master of Engineering,
DATTAKALA’S GROUP OF INSTITUTIONS, SWAMI CHINCHOLI, Tal. Daund,
Dist.Pune 413130, MAHARASHTRA is a bonafide work done by me under supervi-
sion of Dr. S. S. Bere
This submission represents my ideas in my own words and where ideas or words
of others have been included, I have adequately and accurately cited and referenced
the original sources.
I also declare that I have adhered to ethics of academic honesty and integrity
and have not misrepresented or fabricated any data or idea or fact or source in my
submission. I understand that any violation of the above will be a cause for disciplinary
action by the institute and/or the University and can also evoke penal action from the
sources which have thus not been properly cited or from whom proper permission has
not been obtained. This report has not been previously formed the basis for the award
of any degree, diploma or similar title of any other University.

DATTAKALA’S GROUP OF
INSTITUTIONS, SWAMI SOLAVANDE MADHURA

CHINCHOLI, Tal. Daund, Dist. Pune TUKARAM

413130, MAHARASHTRA
3/05/2024
i
Abstract

Object detection stands as a cornerstone in the realm of computer vision, finding

diverse applications spanning from surveillance systems to the development of au-
tonomous vehicles. In this paper, we embark on an exploration of the progressive
strides made in object detection through the lens of machine learning techniques.
Our aim is to delve into the intricacies of methodologies, architectural frameworks,
as well as the merits and challenges inherent in object detection systems powered
by machine learning algorithms. Through an exhaustive literature review and an in-
depth analysis of recent research papers, we unravel the state-of-the-art approaches
that have revolutionized object detection. These approaches not only underscore the
prowess of machine learning but also shed light on the pragmatic implications they
hold for real-world applications. Our investigation traverses through various facets of
object detection methodologies, encompassing the evolution of detection architectures,
advancements in training strategies, and the fusion of multi-modal data for improved
accuracy and robustness. We dissect the intricacies of popular frameworks such
as Faster R-CNN, YOLO, and SSD, elucidating their underlying mechanisms and
comparative performances. Moreover, we dissect the merits accrued from the
adoption of machine learning in object detection systems, including heightened
accuracy, scalability, and adaptability to diverse environments. However, amidst
the advancements, we confront an array of challenges, ranging from data scarcity
and model interpretability to computational complexity and ethical considerations.
Keywords: Object detection, Machine learning, Convolutional neural networks, SSD,
YOLO.

ii
Acknowledgement

I take this opportunity to express my deepest sense of gratitude and sincere thanks to
everyone who helped me to complete this work successfully. I express my sincere
thanks to Dr. A. A. Keste (Principal), Prof. S. D. Salunke (Vice Principal), Dr.
D. B. Hanchate (Head of Department) DEPARTMENT OF COMPUTER ENGI-
NEERING/INFORMATION TECHNOLOGY/AIDS, Dattakala Group of Institutions ,
SWAMI CHINCHOLI, Tal. Daund, Dist. Pune 413130, MAHARASHTRA for
providing me with all the necessary facilities and support.
I would like to express my sincere gratitude to Ms. T. S. Dhage , department of
DEPARTMENT OF COMPUTER ENGINEERING/INFORMATION TECHNOL-
OGY/AIDS, Dattakala Group of Institutions, SWAMI CHINCHOLI, Tal. Daund,
Dist. Pune 413130, MAHARASHTRA for their support and co-operation.
I would like to place on record my sincere gratitude to my seminar guide Dr. S. S.
Bere, Associate Professor, DEPARTMENT OF COMPUTER ENGINEERING/INFORMATION
TECHNOLOGY/AIDS, Dattakala Group of Institutions for the guidance and mentor-
ship throughout the course.
Finally I thank my family, and friends who contributed to the succesful fulfilment
of this seminar work.

Vipula Shamrao Chavan

iii
Contents

Abstract ii

Acknowledgement iii

List of Figures v

List of Symbols vi

1 Introduction 1

2 Litreture Review 3

3 Architecture 5
3.1 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

4 Merits and Demerits of the proposed system 7

5 Challenges 8

6 Conclusion 9

References 10

iv
List of Figures

3.1 Block schematic of object detection [11 . . . . . . . . . . . . . . . . 6

3.2 Basic principle of object detection [12 . . . . . . . . . . . . . . . . . 6

v
List of Symbols

vi
Chapter 1

Introduction

Object detection, a pivotal task in computer vision, involves identifying and precisely
localizing objects within images or video frames. Its applications span various
domains, including security, surveillance, autonomous vehicles, and medical imaging.
Historically, object detection relied on handcrafted features and classifiers, which
often struggled with robustness and generalization. However, the advent of deep
learning, especially convolutional neural networks (CNNs), has catalyzed significant
advancements in this field. Deep learning-based object detection systems have
showcased remarkable performance strides, outperforming traditional methods on
benchmark datasets like COCO (Common Objects in Context) and PASCAL VOC
(Visual Object Classes). These systems harness CNN architectures such as Faster R-
CNN, YOLO (You Only Look Once), and SSD (Single Shot MultiBox Detector) to
achieve accurate and efficient object detection. This report embarks on an exploration
of methodologies, architectures, and recent breakthroughs in object detection using
machine learning techniques. We commence with a meticulous literature review,
analyzing ten recent research papers in the domain of object detection. Subsequently,
we delve into the underlying architecture and methodology of modern object detection
systems. Additionally, we scrutinize the merits and demerits of these systems, along
with the challenges they confront in real-world deployment. In recent years, the
demand for surveillance, detection, and locating systems has significantly increased
across various modern applications. These applications encompass a wide range of
fields including security and traffic surveillance systems, medical applications, and
automated driving systems. As a result, numerous research studies and articles have
emerged, exploring different approaches and methods aimed at developing effective

1
real-time systems. The proliferation of surveillance systems, coupled with the growing
number of CCTV cameras, has presented challenges for operators and control center
personnel. It has become increasingly difficult for individuals to maintain high levels of
efficiency while monitoring multiple cameras simultaneously throughout the day. This
challenge has underscored the need for the development of more effective detection
and tracking systems that operate in real-time. Such systems are crucial not only
for enhancing security measures but also for early detection in various scenarios
such as identifying alien organisms or detecting instances of violent crime. Real-
time detection and tracking systems have proven to be invaluable tools in improving
response times and facilitating prompt intervention in critical situations. These systems
enable operators to efficiently monitor and analyze video feeds, allowing for proactive
measures to be taken as soon as potential threats or anomalies are detected [11-12].
Through this comprehensive examination, we aim to provide insights into the current
landscape of object detection, shedding light on the advancements, challenges, and
potential future directions in this dynamic and rapidly evolving field.

2
Chapter 2

Litreture Review

1. Ren et al. introduced Faster R-CNN, a pioneering deep learning-based object detec-
tion framework that revolutionized the field by achieving state-of-the-art performance.
Faster R-CNN integrates region proposal networks (RPNs) with a convolutional neural
network (CNN) backbone. This integration enables the model to generate region
proposals and perform object detection simultaneously, streamlining the detection
process and significantly improving speed and accuracy [1]. 2. Redmon et al.
introduced You Only Look Once (YOLO), a groundbreaking real-time object detection
system. YOLO processes images in a single pass through a CNN, eliminating the need
for computationally expensive region proposal generation and subsequent refinement
steps. This single-step approach enables YOLO to achieve remarkable speed and
efficiency while maintaining high accuracy in object detection tasks [2]. 3. Liu et
al. presented Single Shot MultiBox Detector (SSD), a versatile framework for object
detection. SSD operates by predicting object classes and bounding box offsets directly
from feature maps at multiple scales. By leveraging feature maps from different layers
of a CNN, SSD achieves robustness and efficiency in detecting objects of varying sizes
and aspect ratios [3]. 4. Lin et al. introduced Mask R-CNN, an extension of Faster
R-CNN that addresses the task of instance segmentation alongside object detection.
Mask R-CNN adds a branch for predicting segmentation masks to the existing Faster
R-CNN architecture. This enables the model to accurately delineate object boundaries,
providing pixel-level segmentation in addition to bounding box detection [4]. 5.
He et al. proposed RetinaNet, a novel object detection framework designed to
tackle the challenge of class imbalance in detection datasets. RetinaNet introduces a
focal loss function that dynamically adjusts the loss weights based on the difficulty

3
of each example. This focal loss prioritizes training on hard examples, thereby
improving the model’s ability to detect objects accurately, especially in the presence
of highly imbalanced datasets [5]. 6. Redmon and Farhadi introduced YOLOv3, an
enhanced version of the YOLO object detection system. YOLOv3 improves upon
its predecessor by introducing multi-scale prediction and feature pyramid networks.
These enhancements enable YOLOv3 to achieve higher accuracy and speed in object
detection tasks across a wide range of scenarios [6]. 7. Law and Deng presented
CornerNet, a novel object detection architecture that adopts a unique approach to
bounding box prediction. CornerNet directly predicts object bounding boxes’ corner
points without relying on anchor boxes, simplifying the detection process and reducing
computational overhead. This approach enables CornerNet to achieve competitive
performance while maintaining simplicity and efficiency [7]. 8. Zhang et al.
introduced Cascade R-CNN, a sophisticated object detection framework designed
to improve detection performance by cascading multiple detection stages. Cascade
R-CNN employs a series of detection stages, each with different Intersection over
Union (IoU) thresholds, to iteratively refine object proposals. This iterative refinement
process enhances the model’s ability to accurately localize objects and reduce false
positives [8]. 9. Zhu et al. proposed CenterNet, a lightweight and efficient object
detection framework that simplifies the detection pipeline by directly predicting object
centers and sizes. Unlike traditional methods that rely on anchor boxes or bounding
box regression, CenterNet eliminates the need for complex post-processing steps,
making it simple yet effective for object detection tasks [9]. 10. Wang et al. presented
EfficientDet, a state-of-the-art object detection framework that achieves remarkable
performance by scaling up a baseline architecture using a compound scaling method.
EfficientDet optimizes model architecture, depth, and width simultaneously to achieve
a balance between efficiency and accuracy. This compound scaling approach enables
EfficientDet to outperform existing object detection models across a wide range of
datasets and computational constraints [10].

4
Chapter 3

Architecture

Modern object detection architectures have evolved to comprise two essential compo-
nents: a backbone network responsible for feature extraction and a detection head for
predicting object attributes such as bounding boxes, class labels, and in some cases,
segmentation masks. The backbone network serves as the foundation for the object
detection system, typically implemented as a Convolutional Neural Network (CNN)
like ResNet or VGG. This network is designed to process input images and extract
hierarchical features that capture different levels of abstraction. For example, early
layers in the backbone network may focus on detecting simple patterns like edges and
textures, while deeper layers learn to represent more complex features such as object
shapes and structures. Once the backbone network has extracted relevant features from
the input image, these features are passed on to the detection head. The detection head
consists of multiple layers of convolutional operations followed by fully connected
layers. Its primary task is to interpret the extracted features and make predictions
about the objects present in the image. Within the detection head, convolutional
layers are typically used to generate spatial information about the objects’ locations
within the image. These layers apply filters across the feature maps obtained from
the backbone network to identify regions of interest where objects are likely to be
present. Following the convolutional layers, fully connected layers are employed to
perform object localization and classification. These layers take the spatial information
obtained from the convolutional layers and refine it to produce precise bounding box
coordinates for each detected object. Additionally, they assign class labels to the
objects based on their visual attributes, enabling the system to differentiate between
different object categories. By combining the feature extraction capabilities of the

5
 Figure 3.1: Block schematic of object detection [11

Figure 3.2: Basic principle of object detection [12

backbone network with the prediction capabilities of the detection head, modern object
detection architectures can accurately and efficiently identify objects within images
across a wide range of applications, from autonomous vehicles to surveillance systems
[12-13]. The Figure 3.1 shows the block schematic of object detection and Figure 3.2
depicts the basic principle of object detection.

3.1 Methodology
1. Data Collection and Annotation: Collecting a diverse dataset of annotated images
containing objects of interest is crucial for training object detection models. 2. Model
Training: Training the object detection model involves optimizing its parameters
using annotated training data. This typically involves minimizing a loss function that
penalizes discrepancies between predicted and ground truth bounding boxes and class
labels. 3. Evaluation: Evaluating the trained model on a separate validation dataset
to assess its performance in terms of accuracy, precision, recall, and computational
efficiency. 4. Deployment: Deploying the trained model in real-world applications,
which may involve optimizing its architecture for inference speed and memory
efficiency [13].

6
Chapter 4

Merits and Demerits of the proposed

system

Merits - High accuracy and efficiency: Modern object detection systems based
on machine learning techniques achieve high accuracy and efficiency, enabling real-
time applications such as autonomous driving and surveillance. - Generalization:
These systems can generalize well to unseen data and adapt to various environmental
conditions and object categories [13].
Demerits: - Data dependency: Object detection models require large amounts
of annotated training data, which may be costly and time-consuming to collect. -
Computational complexity: Training and deploying object detection models can be
computationally intensive, requiring powerful hardware resources [13].

7
Chapter 5

Challenges

1. Data scarcity: Annotated training data for object detection may be scarce, especially
for niche object categories or specific environments. 2. Computational resources:
Training and deploying object detection models require significant computational
resources, limiting their accessibility to researchers and practitioners. 3. Real-world
variability: Object detection models may struggle to generalize to real-world scenarios
with varying lighting conditions, occlusions, and object scales [13].

8
Chapter 6

Conclusion

In conclusion, object detection using machine learning techniques has witnessed

significant advancements in recent years, driven by the proliferation of deep learning
and convolutional neural networks. These advancements have led to highly accurate
and efficient object detection systems capable of real-time performance in various ap-
plications. However, challenges such as data scarcity, computational complexity, and
real-world variability persist and require further research efforts to overcome. Overall,
object detection using machine learning holds immense promise for addressing real-
world challenges and advancing computer vision technologies.

9
References

[1] Ren, S., He, K., Girshick, R., Sun, J. (2015). Faster R-CNN: Towards real-
time object detection with region proposal networks. In Advances in neural
information processing systems (pp. 91-99).

[2] Redmon, J., Divvala, S., Girshick, R., Farhadi, A. (2016). You only look once:
Unified, real-time object detection. In Proceedings of the IEEE conference on
computer vision and pattern recognition (pp. 779-788).

[3] Liu, W., Anguelov, D., Erhan, D., Szegedy, C., Reed, S. (2016). SSD: Single
shot multibox detector. In European conference on computer vision (pp. 21-37).
Springer, Cham..,

[4] . He, K., Gkioxari, G., Dollár, P., Girshick, R. (2017). Mask R-CNN. In
Proceedings of the IEEE international conference on computer vision (pp. 2961-
2969) , ,

[5] .. Lin, T. Y., Goyal, P., Girshick, R., He, K., Dollar, P. (2017). Focal loss for
dense object detection. In Proceedings of the IEEE international conference on
computer vision (pp. 2980-2988). ,

[6] Redmon, J., Farhadi, A. (2018). YOLOv3: An incremental improvement. arXiv

preprint arXiv:1804.02767 ,

[7] Law, H., Deng, J. (2018). Cornernet: Detecting objects as paired keypoints. In
Proceedings of the European conference on computer vision (ECCV) (pp. 734-
750).. ,

[8] Wang, X., Zhang, R., Wu, Y., van der Maaten, L. (2020). EfficientDet: Scalable
and efficient object detection. In Proceedings of the IEEE/CVF Conference on
Computer Vision and Pattern Recognition (pp. 10780-10789).

10
[9] Abdulghafoor, N. H., Abdullah, H. N. (2022). Objects Detection and Tracking
Framework for Different Challenges. Alexandria Engineering Journal, 61(12),
9637-9647. ,

11