Smart Presentation Control by Hand Gestures Using Computer Vision and Google’s Mediapipe

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 07 | July 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 2657
Smart Presentation Control by Hand Gestures Using
Computer Vision and Google’s Mediapipe
Hajeera Khanum1, Dr. Pramod H B2
1M.Tech Student, Dept. of Computer Science Engineering, Rajeev Institute of Technology, Hassan, Karnataka, India
2Associate Professor, Dept. of Computer Science Engineering, Rajeev Institute of Technology, Hassan, India
---------------------------------------------------------------------***---------------------------------------------------------------------
ABSTRACT - Using hand gestures as the system's input to
control presentation, we are constructing a presentation
controller in this paper. The OpenCV module is mostly utilised
in this implementation to control the gestures. MediaPipe is a
machine learning framework with a hand gesture detection
technology that is available today. This system primarily
employs a web camera to record or capturephotosandvideos,
and this application regulates the system'spresentation based
on the input. The primary purpose of the system is to change
its presentation slides, I also had access to a pointer that
allowed me draw on slides, in addition to that erase. To
operate a computer's fundamental functions, such as
presentation control, we may utilise hand gestures. People
won't have to acquire the often burdensome machine-like
abilities as a result. These hand gesture systems offer a
modern, inventive, and natural means of nonverbal
communication. These systems are used widely in human-
computer interaction. This project's purpose is to discuss a
presentation control system based on hand gesture detection
and hand gesture recognition. A high resolution camera is
used in this system to recognise the user's gestures as input.
The main objective of hand gesture recognition is to developa
system that can recognise human hand gestures and use that
information to control a presentation. With real-time gesture
recognition, a specific user can control a computer by making
hand gestures in front of a system camera that is connected to
a computer. With the aid of OpenCVPythonandMediaPipe, we
are creating a hand gesture presentation control system in
this project. Without using a keyboard or mouse, this system
can be operated with hand gestures.
Keywords: OpenCV, MediaPipe, Hand GestureRecognition,
MachineLearning, presentationcontroller,HumanComputer
Interaction
1. INTRODUCTION
Industry that is 4.0, also known as the Fourth
Industrial revolution, calls for automation and
computerization, which are realised through the
convergence of various physical and digital technologies
such as sensors, embedded systems, Artificial Intelligence
(AI), Cloud Computing, Big Data, Adaptive Robotics,
Augmented Reality, Additive Manufacturing (AM), and
Internet of Things [2]. The increased interconnectedness of
digital technology’s make it essential for us tocarryoutdaily
activities likeworking,shopping,communicating,having fun,
and even looking for information and news [3]. The use of
technologies and improvements in human-machine
interaction allow people to identify, communicate, and
engage with one another using a wide variety of gestures.
The gesture is a type of nonverbal communication
or nonvocal communication that makes use of the body's
movement to express a specific message. The hand or face
are the most frequently used portions of the body [4]. The
research has gravitated toward the new sort of Human-
Computer Interaction (HCI) known as gesture-based
interaction, which Krueger launched in the middle of the
1970s. Building application interfaces with controlling each
human body part to communicate organically is a major
focus of study in the field of human-computer interaction
(HCI), with hands serving as the most practical alternativeto
other interaction tools given their capabilities [5].
Recognizing hand movements using Human-
Computer-Interaction (HCI) might aid in achieving the
necessary ease and naturalness [6]. Hand gestures servethe
purpose of communicating information when engaging with
other individuals. encompassing both basicandcomplicated
hand motions. For instance, we can point with our hands
towards an item or at individuals, or we can convey basic
hand shapes or motions using manual articulations in
conjunction with sign languages' well-known syntax and
lexicon. Therefore, employing hand gestures as a tool and
integrating them with computers might enable more
intuitivecommunicationbetweenindividuals[6].Tosimplify
things to anybody thus create Artificial Intelligence (AI)
based apps, various frameworks or libraries have been
developed for hand gesture detection. MediaPipe is one of
them. For the purpose of employing machine learning
techniques like Face Detection, Iris, Pose, Hands, , Hair
segmentation Holistic,Boxtracking,Objectdetection,Instant
Motion Tracking, Face Mesh, KIFT,and Objection,Google has
created the mediapipe framework. Few benefits for
employing mediapipe framework's showcases include
helping programmer concentrate on model and algorithm
creation for application and supporting application’s
environment via result repeatable allover multiple
architecture and gadgets. To conduct different activities,
such as seeking ahead and backwardthroughslides,drawing
and erasing in a presentation, the project employs hand
gestures, often no of raised fingers inside the region of
interest.

The difficult component of this system is
background movies or pictures that are recorded or
captured while taking inputs, suchashandgesturesfrom the
user. Lightning may also sometimes affect the quality of the
input obtained, whichmakes itdifficulttorecognisemotions.
Segmentation is the process of identifying a linkedarea of an
image that has certain characteristics like colour, intensity,
and a relationship between pixels, or pattern. Additionally,
have utilised some significant packages, like mediapipe,
tensorflow, numpy, and opencv-python.
2. EXISTING SYSTEM
The author has developed an ANN application used
for classificationandgesture recognition,GestureRecognition
Utilizing Accelerometer. The Wii remote, which rotatesin the
X, Y, and Z directions, is essentially employed in this system.
The author has utilised two tiers to construct the system in
order to reduce the cost and memory requirements. Theuser
is verified for gesture recognition at the first level. Author's
preferred approach for gesture recognitionisaccelerometer-
based.
Following that, system signals are analysed at the
second level utilising automata to recognisegestures(Fuzzy).
The Fast Fourier technique and k means are then used to
normalise the data. The accuracy of recognition has now
increased to 95%.
Recognition of Hand Gestures Using Hidden Markov
Models - The author of this work has developed a system that
uses dynamic hand movements to detect the digits 0 through
9. In this work, the author employed two stages.
Preprocessing is done in the first phase, while categorization
is done in the second. There are essentially two categories of
gestures. both Link gestures and Key motions. The key
gesture and the link gestures are employed in continuous
gestures for the goal of spotting. Discrete Hidden Markov
Model (DHMM) is employed for classification in this work.
The Baum-Welch algorithm is used to train this DHMM.HMM
has an average recognition rate range of 93.84 to 97.34%.
The author has employed inexpensive cameras to
keep costs down for the consumers. Robust Part-Based Hand
Gesture Recognition Using Kinect Sensor. Although a kinect
sensor's resolutions lower than that of other cameras, it is
nevertheless capable of detecting andcapturinglarge pictures
and objects. Only the fingers, not the entire hand, are paired
with FEMD to deal with the loud hand movements. This
technology performs flawlesslyand effectivelyinuncontrolled
settings. The experimental result yieldsanaccuracyof93.2%.
3. RELATED WORK
3.1 Hand Gesture Recognition
Computer science's key field of gesture recognition
develops technology that tries to understand human motions
so that anybody may use basic gesturestocommunicatewith
a device without touching it directly. Gesture recognition is
the process of tracking gestures, representing them, and
translating them into a specific instruction[8]. The goal of
hand gesture recognition is to identify from explicit hand
movements as input, then process these motions
representation for devices by mapping as output. The
software sub-tree includes an image recognition tool that
accepts video feeds as input. Using third-party tools like
OpenCV, it is possible to locate MediaPipe objects that are
visible in the camera's field of view.
Fig -1: Software Sub-Tree
There are three hand gesture recognition techniques that
may be found in various types of literature:
Machine Learning Techniques: The condensation
method, PCA, HMM [9][10][11][12], sophisticated particle
filtering, and stochastic processes and approaches based on
statistical models for dynamical gestures produced the
resultant output.
Algorithm perspectives: collection of manually
defined, encoded constraints and requirements for defining
gestures in dynamic gestures. Galveia [13] used a 3rd-degree
numerical solution (construct a 3rd-degree polynomial
equation, detection, decreased complexity for equations, and
comparative handled in gestures libraries) to ascertain the
dynamic element of the hand movements.
Rule-based methods: Appropriate for dynamic
movement alternatively fixed gestures, thatareinputswitha
pre-encoded body of norms [5]. The characteristics of input
movements are retrieved, and they are evaluated to the
encoded principles that regulate the flow of the motions that
are recognised Advances in Engineering Research, volume
207 102. synchronization between rules-based gestures and

input that is accepted as recognised gestures upon output
[13].
3.2 MediaPipe Framework
For the recognition of hand gestures, there are
several machine learning frameworks and tools available
today. MediaPipe is among them. The MediaPipe is just a
framework created to providemanufacturing-ready machine
learning, which requires build infrastructure to execute
inference above any sort of sensory information and has
released code to go along with scientific work [7]. The
function of the media processor model, inference model and
the data manipulation are all drawn from a perceptual
pipeline in MediaPipe [14]. Other machine learning
systems likeOpenCV4.0, Tensorflow,PyTorch,MXNet, CNTK,
also employ graphs of computations.
Fig -2: Overview of the Hand Perception Pathway
Two hand gesture recognitionmodelsareimplementedby the
MediaPipe of Figure 2 as follows [14]:
1. A palm detecting described the basic the acquired
picture and transforms the images with such an
aligned object of a hand.
2. The hand landmark modelprocessesanimagewitha
clipped bounding box and produces 3D hands key
points just on hands.
3. The gesture recognition system that organises 3D
hands key point into a distinct set of motions after
classifying them.
3.3 Palm Detection Pattern
The BlazePalm first palm detector was implemented
inside the MediaPipe framework. The detectionofthe handis
a difficult process. In order to model the palm using square
bounding boxes to prevent other aspect ratios and lowering
the amount of hooks by the ratio of 3–5, non-maximum
reduction method must first be trained on the palm rather
than the hand detector. Finally, limit the focus loss during
training with help from a huge number of anchoring caused
by the high exist in a wide using encoder-decoder of image
retrieval that is employed for larger scene context-awareness
even for tiny objects.
3.4 Hand Landmark
Accomplishes accurate crucial point clustering of 21 main
points with only a 3D touch coordinates that is done within
the identified hand areas and immediately generates the
coordinates predictor that is a representation of hand
landmarks within MediaPipe [15][16].
Fig -3: MediaPipe Hand Landmark [18]
Every touch of a landmark had already coordinate is
constituted of x, y, and z in which x and y have been adjusted
to [0.0, 1.0] besides image width and length, while z
portrayal the complexity of landmark. A depth of the
ancestral landmark, which is located at the wrist. The value
decreases the more away the landmark is from the camera.
4. SYSTEM ARCHITECTURE AND METHODOLOGY
The code for this project was created in the python
language utilising the Opensv and NumPy packages. In this
work, the libraries that will be utilised for further input and
output processing are initially imported. MediaPipe,
OpenCV and numpy are the libraries that are utilised in this
project and that need to be imported. Videoinputscomefrom
in out main camera.To recognise the video as input from our
camera, mediapipe is now being utilised, and the
mphand.hands module is being used to detect the gesture.
Then, in order to access the presentation, weutilised pointer.
The input picture must next be converted to an RGB image to
finish the input processing. Then it's your chance to specify
the thumb and finger points in input. Numpy is utilised to
transform this process needed output. presentation is
handled using the hand range in this procedure. The Python
language's NumPy library is essential for computing. It
includes a variety of elements:

1. effective N-dimensional array
2. Object broadcast and C integration tools
3. Capability for the Fourier analysis and pseudo
random
4.1 Identifying Hand Gesture by MediaPipe
Across platforms including Android, iOS, the web,
edge devices, and many applicable MLpipelines,MediaPipeis
a module for processing video, audio, and various sorts of
related data. With the use of this module, a variety of tasks
may be completed
Fig -4: System Architecture
In our project, we utilised it to identify hand gestures and
extract input for:
1. Multi-hand Monitoring
2. Facial Recognition
3. Segmentation
4. object tracking
5. object detection
In order to generate a better result, we have implemented a
Hand Gestures Recognition System. The webcam is turned
on while the software is running, and the kind of gesture
used to detect the shape of the hand and give us the desired
output is static. This project uses the curve of the hand to
regulate loudness. The system receives input, captures the
item, detects it, and then recognises hand gestures.
4.2 NumPy
A Python package called Open CV addresses the
problem of PC vision. It is utilised for face detection, whichis
carried out utilising machine learning. It is a highly
significant library that is used in several applications to
identify various frames and detect faces. It also supports a
number of programming languages. Additionally, it carries
out motion and object detection. It may be used to recognise
the faces of animals and supports a different operating
system.
4.3 TensorFlow
Google developed TensorFlow, a framework that
enables programmers to use "novel optimizations and
training algorithms" for defining, developing, and using
various machine learning models. The machine learning
algorithms of TensorFlow may be thought of as directed or
computational graphs. Each node in such a network denotes
an operation, and the edges (tensors) indicate the data that
moves between the operations. In order to generate a better
result, we have implemented a Hand Gestures Recognition
System. TensorFlow's model was modified in the library to
only return necessary points on the body.PoseNetisused by
the pose estimation programme to locate joints on the
human body. The TensorFlow posture estimation library
contains PoseNet, a pre-trained model that uses computer
vision to predict a person's bodily joints [. Seven jointsinthe
body are given coordinates with numbers ranging from 0 to
6.
TensorFlow has demonstrated the ability to offer
solutions for object recognition using photos and may be
used to train huge datasets to recognise specific things.
TensorFlow also features a library that the user
might just save or reload as needed. This enables users to
save the checkpoints with the highest evaluation score and
makes it reusable for unsupervised learning or model fine-
tuning.
5. RESULT
Fig -5: Hand Gesture to move on to the next slide

Fig -6: Hand Gesture for going back to previous slide
Fig -7: Getting a pointer on slide
Fig -8: Draw using that pointer
Fig -9: Erase the drawing on slide
6. CONCLUSION
This project showcases a programme that enables
hand gestures as a practical and simple method of software
control. A gesture-based presentation controller doesn't
need any special markers, and it can be used in real life on
basic PCs with inexpensive cameras since it doesn't need
particularly high quality cameras to recognise or record the
hand movements. The method keeps track ofthelocations of
each hand's index finger and counter tips. This kind of
system's primary goal is to essentially automate system
components so that they are easy to control. As a result, we
have employed this method to make the system simpler to
control with the aid of these applications in order to make it
realistic.
REFERENCES
[1] Research Gate, Google.
[2] Cevikcan, Ustunug A, Industry 4.0: ManagingTheDigital
Transformation, Springer Series in Advanced
Manufacturing, Switzerland. 2018. the following DOI:
10.1007/978-3-319-57870-5.
[3] Pantic M, Nijholt A, Pentland A, Huanag TS, Human-
Centered Intelligent Human-Computer Interaction
(HCI2): How Far We From Attaining It?,International
Jounal of Autonomous and Adaptive Communications
Systems (IJAACS), vol.1 no.2, 2008. pp 168-187. DOI:
10.1504/IJAACS.2008.019799.
[4] Hamed Al-Saedi A.K, Hassin Al-Asadi A, Survey of Hand
Gesture Recognition System. IOP Conferences Series:
Journal of Physics: Conferences Series 1294 042003.
2019.
[5] Z.Ren, J.Meng, Yuan J.Depth Camera BasedHandGesture
Regconition and its Application in Human-Computer-
Interaction. In Processing of the 2011 8th International
Conference on Information, Communication and Signal
Processing (ICICS). Singapore. 2011.
[6] S.Rautaray S, Agrawal A. Vision Based Hand Gesture
Recognition for Human Computer Interaction:ASurvey.
Springer Artificial Intelligence Review. 2012. DOI:
https://doi.org/10.1007/s10462-012-9356-9.
[7] Lugaresi C, Tang J, Nash H, McClanahan C, et al.
MediaPipe: A Framework for Building Perception
Pipelines. Google Research. 2019.
https://arxiv.org/abs/2006.10214.
[8] Z.Xu, et.al, Hand Gesture Recognition and Virtual Game
Control Based on 3D AccelerometerandEMGSensors,In
Processing og IUI’09, 2009, pp 401-406.

[9] Lee H, Kim J. An HMM-Based ThresholdModel Approach
for Gesture Recognition. IEEE Trans on PAMI vol.21.
1999. pp 961-973.
[10] Wilson A, Bobick A, Parametric Hidden Markov
Models for Gesture Recognition. IEEE Trans. On PAMI
vol.21, 1999. pp.884-900.
[11] Wu Xiayou, An Intelligent Interactive System Based
on Hand Gesture Recognition Algorithm and Kinect, In
5th International Symposium on Computational
Intelligence and Design.2012.
[12] Wang Y, Kinect Based Dynamic Hand Gesture
Recognition Algorithm Research, In 4th International
Conference on Intelligent Human Machine System and
Cybernetics. 2012.
[13] Galveia B, Cardoso T, Rybarczyk, Adding Value to
The Kinect SDK Creating a Gesture Library, 2014.
[14]Lugaresi C, Tang J, Nash H et.al, MediaPipe: A
Framework for Perceiving and Processing Reality.
Google Research. 2019.
[15]Zhag F, Bazarevsky, Vakunov A et.al, MediaPipe Hands:
On – Device Real Time Hand Tracking, Google Research.
USA. 2020. https://arxiv.org/pdf/2006.10214.pdf
[16]MediaPipe: On-Device, Real Time Hand Tracking, In
https://ai.googleblog.com/2019/08/on-device real-
time-hand-tracking-with.html. 2019. Access 2021.
[17]Grishchenko I, Bazarevsky V, MediaPipe Holositic –
Simultaneoue Face, HandandPosePredictiononDevice,
Google Research, USA, 2020,
https://ai.googleblog.com/2020/12/mediapipe holistic
-simultaneous-face.html, Access 2021.
[18]MediaPipe Github:
https://google.github.io/mediapipe/solutions/hands.
Access 2021.

Smart Presentation Control by Hand Gestures Using Computer Vision and Google’s Mediapipe

More Related Content

What's hot (20)

Similar to Smart Presentation Control by Hand Gestures Using Computer Vision and Google’s Mediapipe (20)

More from IRJET Journal (20)

Recently uploaded (20)

Smart Presentation Control by Hand Gestures Using Computer Vision and Google’s Mediapipe