Olympic Sports Recognition

Chitra Lakhani, Vikas Janu

JK Lakshmipat University Jaipur 302026 India

ARTICLE INFO ABSTRACT

Keywords: This is a brief report about the process used for the image recognition of any Olympic sports. First,
Image processing there is a brief introduction to the report, followed by a background on big data and image recognition.
Neural Network Then, the methodology is explained that is used to complete this project and achieve the objective of
Image Recognition this project. The objective of this report is to give a report on the project which is the recognition of
Big data any Olympic sport from any image using image processing.

1. Introduction real-time, to produce insights that can be used to inform

business choices.
Sports recognition using image processing is a field that Building scalable and dependable systems that can man-
involves using computer algorithms to identify and under-
age the large volume, diversity, and velocity of data that
stand motion-related images or videos. It is an important
enterprises generate is the main objective of big data en-
element of research and development and has received great gineering. Data modelling, data integration, data storage,
attention in recent years due to its many uses. data processing, and data analysis must all be thoroughly
The main idea of image recognition is to use appropriate understood to do this. Programming languages like Java,
information from pictures or videos, such as players’ posi- Python, and Scala as well as software engineering techniques
tions, balls, and other objects, and use this information to
like version control, testing, and debugging must all be
determine what is happening during the game. This may
mastered by big data engineers.
include tasks such as tracking players, finding the ball, and Big Data engineers are in high demand, and this trend
recognizing the sports. is anticipated to continue in the coming years. The demand
Use a variety of computer vision, such as object de- for knowledgeable Big Data Engineers will only rise as more
tection, segmentation, and classification, to perform these and more firms rely on data-driven insights to make choices.
tasks. These techniques involve visual analysis of images or
videos and use them to identify and track items of interest. 2.2. Image Processing
It has many applications such as sports recognition using The alteration of images is the focus of the branch of
functional images, sports advertising, player analysis and digital signal processing known as image processing. In
sports education. For example, it can be used to follow the order to obtain valuable information from digital photos, it
movements of the players and the ball during football and involves techniques for image enhancement, analysis, and
to give detailed information about the performances of the interpretation. Image processing has a wide range of uses
players. It can also be used to help coaches and players in industries like entertainment, engineering, security, and
identify areas of improvement and develop better training medical. Picture enhancement, picture restoration, image
strategies. segmentation, and image identification are some of the fre-
Overall, the experience of recognizing sports using pic- quent tasks carried out by image processing algorithms.
tures is exciting and evolving rapidly with the potential to Image enhancement is the process of enhancing an im-
change the way we understand and analyse images. age’s visual quality, such as by reducing noise, sharpening
the edges, or modifying the brightness and contrast. Contrar-
2. Related Work ily, image restoration seeks to improve deteriorated images
caused by noise, blur, or other elements.
2.1. Big Data Engineering In order to detect objects of interest, image segmentation
Big Data Engineering is the practise of planning, con- involves separating an image into several areas. This method
structing, evaluating, and maintaining systems that can man- is frequently applied in medical imaging to find lesions
age significant amounts of data. The expansion in the volume or tumours. Automatically identifying objects or patterns
of data produced by companies, people, and organisations inside a picture is known as image recognition. Examples
has given rise to this field. The infrastructure needed to store, of this include face recognition and object detection in self-
process, and analyse this data is built by big data engineers. driving automobiles.
Many different technologies and techniques are included MATLAB, Python, and C++ are just a few of the many
in the field of big data engineering, such as distributed computer languages that can be used to develop image
computing systems, cloud computing, Hadoop, spark, and processing methods. Additionally, there are numerous spe-
no-SQL databases. With the help of these technologies, cialized software programmes and libraries for image pro-
enormous amounts of data can be processed, frequently in cessing, including ImageJ, OpenCV, and scikit-image.
ORCID (s):

Alok Kumar et al.: Preprint submitted to Elsevier Page 1 of 4

 CS1312: Project Report

In conclusion, the field of image processing is crucial

to many facets of contemporary life. Its methodologies are
constantly changing in line with developments in artificial
intelligence and computation, and its uses span from surveil-
lance and entertainment to medical diagnosis and treatment.

3. Olympic Sports Recognition

The process of recognising objects, persons, or other
things in digital photos or movies is known as image recog-
nition. In addition to security, medical imaging, driverless
vehicles, and augmented reality, it is a significant application
of computer vision and machine learning.
Following are the steps followed for image recognition
process:
1. Data gathering and preparation: Gathering and getting
ready the data is the initial phase. In order to do this, a sizable
dataset of photos must be gathered, each of which contains
examples of the objects or entities the system is supposed
to be able to identify. To make sure the system can use the
dataset to learn from it efficiently, it must be appropriately
curated and labelled.
2. Feature extraction: The system extracts feature from
the photos in this step that are important for recognition. A
feature may be simple, like colour or texture, or it may be mins to train, and will take longer with more input data to
complicated, like shapes and patterns. The attributes chosen train on and more epochs to run.
depend on the application in question and how difficult the
recognition task is to complete.
3. Training: To understand the connections between the Appendix
features and the recognised objects or entities, the system is
taught using machine learning methods, such as deep neural
A. About the Data
networks. The system is fed labelled data during the training The type of data that is used is images. We will use the
phase, and the model’s parameters are changed to improve 100s of images from each of the sports played at Olympics to
performance. recognise which sports is which. The data is in images and
4. Testing and validation: Following training, the system the processing will be done on these images. The csv file
is put to the test on a different collection of photos to gauge contains the location of these images and the classification
how well it performs. To determine the system’s effective- for them. We will train the model on the given data and then
ness, accuracy and precision are measured, and the findings test that model for accuracy to recognise the sports.
are contrasted with the actual data.
5. Deployment and refinement: The system can be im-
B. Data Gathering
plemented in the actual world following testing and vali-
dation. The system might need to be improved over time For building this model, the requirements are 100s of
nevertheless as it runs across new problems and data. This images of all the Olympic games that are played. For this
entails tracking the system’s performance over time, finding dataset, we found 61 such games and have collected ap-
potential improvement areas, and updating the model and the proximately 100 images for each of these games. The data
data as necessary. collection was ethical and copyright free since the images
were collected from free and no-copyright resources.

4. Evaluation and Results

The result of the model built in this project is given
C. Data Preprocessing
below: The data is loaded for processing first. This is where we
As the number of epochs increase, accuracy of the model will start working in Jupyter notebook. We can view the data
increases, and the loss is decreasing. Nevertheless, there is as follows to see if it is accessible:
still only 60% accuracy, approximately, and there is still area After that, we can load by accessing the directory di-
for growth for the model. We can increase the number of rectly through the notebook with os library that is imported
inputs to the model to make it more precise and increase the as follows:
number of epochs. Although, that would take longer of the Here, we have adjusted the size of all the images in
model to train. This model took approximately 900s or 15 order to make it easy for them to fit the model and given all

Alok Kumar et al.: Preprint submitted to Elsevier Page 2 of 4

 CS1312: Project Report

the images the same size, 100x100, using the cv2.resize()

function from the cv2 library. To confirm that the data has
been imported:
the input data, allowing them to capture complex patterns
and structures that are difficult to identify using traditional
machine learning techniques.

Working on CNN
The basic building block of a CNN is the convolutional
layer. In this layer, a set of filters (also known as kernels)
are applied to the input image, with each filter learning to
detect a specific feature or pattern. The filters are typically
small in size (e.g. 3x3 or 5x5), and they slide over the input
image one pixel at a time, producing a 2D activation map
(also known as a feature map) that highlights the regions of
the input image that correspond to the learned feature.
The output of a convolutional layer is then passed
D. Data Augnmentation through a non-linear activation function, such as ReLU
Data augmentation refers to the technique of artificially (Rectified Linear Unit), which introduces non-linearity into
increasing the size of a training dataset by applying vari- the network and allows it to capture more complex patterns.
ous transformations to the existing data. The goal of data The resulting feature maps are then fed into a pooling layer,
augmentation is to improve the performance of a machine which reduces the spatial size of the feature maps by taking
learning model by increasing the diversity of the training the maximum or average value over small regions (e.g. 2x2)
data, which helps the model to generalize better to new, in each feature map. This pooling operation helps to make
unseen data. Some common features are: the network more robust to small variations in the input
In addition to image data, data augmentation can also data, and reduces the number of parameters in the network,
be applied to other types of data, such as text or audio making it computationally more efficient.
data. For example, in natural language processing (NLP), The output of the pooling layer is then passed through
data augmentation techniques can include paraphrasing or another set of convolutional and pooling layers, creating a
altering the order of sentences in a document. By increasing hierarchy of feature maps that capture increasingly complex
the size and diversity of the training data, data augmentation patterns and structures in the input data. Finally, the output
can help to prevent overfitting and improve the robustness of the last convolutional layer is flattened into a 1D vector
and generalization ability of the model. In this model, we and passed through one or more fully connected layers,
have used ImageDataGenerator to apply augmentation with which perform a non-linear transformation of the features
few of the above mentioned features. and produce the final output of the network.
During training, the parameters (i.e. weights and bi-
ases) of the network are learned using backpropagation,
E. Model Building a gradient-based optimization algorithm that adjusts the
The model used for this project is one the most common parameters to minimize the difference between the network’s
models used for images recognition, CNN (Convolutional output and the desired output (e.g. a class label for an image).
Neural Network). The learning process involves iteratively feeding training
Convolutional Neural Networks (CNNs) are a type of examples through the network, computing the error between
artificial neural network that are commonly used for image the predicted output and the true output, and updating the
recognition and computer vision tasks. CNNs are designed parameters of the network using the gradients of the error
to automatically learn spatial hierarchies of features from with respect to the parameters.

Alok Kumar et al.: Preprint submitted to Elsevier Page 3 of 4

 CS1312: Project Report

2020BTECHCSE024
JK Lakshmipat University, Jaipur

Vikas Janu
2020BTECHCSE083
JK Lakshmipat University, Jaipur

E.1. Model Building in Notebook

First we will build the model and then fit the generated
data in it.
Model training:

F. Source Code
https://github.com/chitra-sl/Big-data-Project

G. IPR Certificate
We, Chitra S. Lakhani and Vikas Janu, with this certify
that the project work submitted by us, entitled Olympic
Sports Recognition, to our supervisors, Dr. Alok Kumar and
Dr. Utsav Upadhyay, in partial fulfilment of the requirements
for the course is a Bonafede work carried out by us and
has not been previously submitted to any other course. We
further certify that no part of this work shall be published,
reproduced, or distributed in any form without the prior
permission of our supervisors. We understand that any such
unauthorized use of the project work may be considered a
violation of academic ethics and result in severe penalties.
We also affirm that the project work has been carried out
under the ethical standards and guidelines set forth by our
supervisors. We acknowledge that our supervisor has the
right to make any modifications or revisions to the project
work that may be deemed necessary. We also agree to abide
by any additional terms and conditions as stipulated by our
supervisor.

Date: 20th May, 2023

Chitra Lakhani

Alok Kumar et al.: Preprint submitted to Elsevier Page 4 of 4