International Journal of Scientific Research in Engineering and Management (IJSREM)

Volume: 08 Issue: 12 | Dec - 2024 SJIF Rating: 8.448 ISSN: 2582-3930

Gesture Recognition for Interactive Presentation Control: A Deep

Learning andEdge Computing Approach on Raspberry Pi

Prashanth N

Assistant Professor
Dept of Artificial Intelligence and Machine Learning Vidyavardhaka College of Engineering, Gokulum, Mysore

K T Anchan Poovanna1, Gaurav Raj2, Preetham S3, Qurrath Ul Ayen4

BE Students
Dept of Artificial Intelligence and Machine Learning Vidyavardhaka College of Engineering, Gokulum, Mysore

Abstract Keywords

Gesture Recognition Human-Computer Inter-

Gesture recognition is one of the important tech- action (HCI) Edge Computing Deep Learning
nologies for intuitive HCI applications, for ex- Real-Time Gesture Detection
ample, in interactive presentations. This re- view
focuses on recent work on gesture recog- nition
systems, deep learning-based methods for 1 Introduction
presentation control by hand gestures, on edge
As digital interaction develops, gesture recogni-
devices like the Raspberry Pi. This work de-
tion has increasingly come to be seen as a tech-
scribes both static and dynamic frameworks for
nology that allows for intuitive HCI. The main
gesture recognition and human-machine interac-
problem in this research area is therefore how to
tion techniques based on deep learning.Recent
develop systems capable of real-time gesture
studies show the usability of real-time hand
recognition for interactive presentation control on
gesture recognition using Python-based systems via
low-cost, portable edge devices such as the
Raspberry Pi, stressing portability and ef-
Raspberry Pi. The system will accomplish a
ficiency in solutions. The review discusses var-
responsive, hands-free presentation system that can
ious deep learning frameworks such as Tensor-
interpret static and dynamic hand gestures, thus
Flow, which can be used in image preprocess-
being a potential substitute or complement to
ing and model training to enhance the accuracy of
standard input devices. This is done in the context
gesture detection. Techniques cover wear- able
of availabilities in ML, edge computing, and
devices to computer vision and demon- strate the
computer vision research lines, aiming for
flexibility of gesture recognition in a variety of
technology concerning gesture recognition into the
hardware platforms. Advances in edge computing
road towards greater accessibility, greater
enable complex gesture recog- nition on low-
efficiency, and more exigent real-time applica-
power devices, with reduced la- tency, and
tions.
enables improved accessibility.The re- view
addresses challenges such as gesture type It is a very demanding field of research in
distinction, responsiveness, and environmental robotics, virtual reality, and smart technology for
variability in developing reliable gesture-based HCI. Deep models had achieved highly ac- curate
systems. In this regard, it emphasizes the Rasp- classification for complex gestures. Most of the
berry Pi as an edge computing solution for in- systems based on gestures are, however, rather
teractive presentation control in HCI. very powerful computing usually cloud- based
infrastructures that add latency and de-

© 2024, IJSREM | www.ijsrem.com DOI: 10.55041/IJSREM39636 | Page 1

 International Journal of Scientific Research in Engineering and Management (IJSREM)
Volume: 08 Issue: 12 | Dec - 2024 SJIF Rating: 8.448 ISSN: 2582-3930

Key research questions include:

grade data privacy. It focuses on enabling ges-
ture recognition in edge devices to allow real- 1. Ways of improving deep-learning models to
time processing, minimize the number of net- respond to static and dynamic gestures using a
work transmissions, and thereby deliver cost- Raspberry Pi What pre-processing shall be
effective practical solutions for classrooms and performed to get reliable results with such varied
conference rooms. lighting conditions?
A relevant literature review for this research 2. How to minimize latency in order to enable
collates existing knowledge on gestures recog- real-time control without pause?
nition and identifies the latest advancements of The literature review will touch on gesture
deep learning and edge computing for low power recognition frameworks, especially deep learn- ing
devices. In this regard, such a review would and image preprocessing, targeted for low- power
encompass seminal methodologies as well as devices. Development and analysis tools will
relevant frameworks that entail TensorFlow comprise TensorFlow, OpenCV, and Medi- aPipe.
which can efficiently realize gesture detection in Since the research is specifically focused on
resource-constrained devices given the chal- lenges gesture-based control, voice commands and other
of differences in gesture types, variability in modalities of HCI such as speech interac- tion
environments, and latency reduction. have been omitted from the study for sim- plicity.
This research is important because it adds This research will lead to an approach that can be
interactive presentation control, allowing us to practically applied to edge comput- ing to give
highlight our work as more engaging and ac- gesture-based HCI access to perfor- mance and
cessible. Our findings could be significant in a efficiency across a variety of environ- ments.
broader HCI landscape, depicting the oppor-
tunity to combine the potential of deep learn- ing 3 Literature Review
with edge computing in creating responsive
gesture-based systems that can be deployed on 3.1 Python-based Raspberry
affordable hardware, such as Raspberry Pi. Pi for Hand Gesture Recognition[1]
3.1.1 Methodologies
2 Background & Significance The paper describes a hand gesture recognition
system for controlling a mobile robot through
Gesture recognition has become the transform- ing Raspberry Pi in which video frames are captured and
element of human-computer interaction, in as processed using Python and OpenCV, ap- plying
much as it allows intuitive control of the digital operations like Gaussian blur, threshold- ing,
system even hands-free. This work ad- dresses contour detection to differentiate the hand shapes
one of those challenges: developing real- time and the positions of fingers. The ap- plied gestures
gesture control systems specifically tailored for are commands like forward and backward
presentation control on a low-cost edge de- vice, movements that will execute through GPIO pins
the Raspberry Pi. It provides an optimally smooth of the Raspberry Pi.
and hands-free navigation using deep learning
techniques on a fully portable, resource- 3.1.2 Limitations
constrained platform.
These advantages include, to be relatively spe- The system has limitations in that it reads only
cific, lower latency and better privacy. However, basic finger-count gestures, limiting its applica-
the integration of deep learning on such devices tions. It performs well in consistent lighting and
poses serious challenges in the form of computa- simple backgrounds and is less adequate under
tional constraints and real-time requirements. varied environments. Classical image processing
techniques applied to this system limit the accu- racy
This research is important because gesture- based
for complex gestures, which deep learning may be
control can redefine user engagement in
more effective at resolving. The absence of
presentations and make them more interactive
feedback, coupled with the slow processing of the
and accessible. Using a low-cost edge device en-
Raspberry Pi, also leads to latency in real- time
ables these applications to go beyond high-end
response.
systems into education, small businesses, and re-
mote areas where affordability and portability
become key.

© 2024, IJSREM | www.ijsrem.com DOI: 10.55041/IJSREM39636 | Page 2

 International Journal of Scientific Research in Engineering and Management (IJSREM)
Volume: 08 Issue: 12 | Dec - 2024 SJIF Rating: 8.448 ISSN: 2582-3930

3.1.3 Learnings diapipe’s solutions.hands module detects hand

landmarks, while Euclidean distance calcula- tions
This study’s methods can be used for a gesture-
identify specific gestures. Machine learning models,
controlled presentation system with a Raspberry Pi
such as TensorFlow models trained on custom
camera. Pre-processing steps are better done with
datasets, support applications like sign language
Gaussian blur and contour detection, and with
recognition, with gestures mapped to specific
the upgrade of hardware or incorporation of a
functions through a Tkinter interface.
USB accelerator, latency will be minimized,
aiming for better real-time gesture recognition in
presentations.
Limitations

3.2 A Deep Learning Framework for The system encounters challenges in maintain- ing
Recognizing Both Static and Dynamic accuracy under non-ideal lighting and is con-
Gestures[2] strained by a reliance on predefined gestures.
Mediapipe’s performance varies under certain
3.2.1 Methodologies conditions, which can impact gesture recognition
accuracy and usability in diverse environments.
taDNet uses pose-driven spatial attention with
the use of OpenPose for extracting pose and
3.3.2 Learnings
depth estimation. The former is captured by a
CNN where LSTMs are utilized for modeling the This paper provides useful methods for gesture
temporal aspects of dynamic gestures. Fur- detection, particularly through Mediapipe and
thermore, cropping RGB frames around hand OpenCV, which you can use for real-time hand
regions fosters gesture detection in a sensor-less tracking in your presentation system. Imple-
scenario while normalizing joint velocities and menting similar gesture-mapping techniques al- lows
accelerations improve accuracy. you to customize commands for slides and
multimedia control. Leveraging TensorFlow Lite
models on Raspberry Pi can enhance gesture
3.2.2 Limitations
recognition speed, while data augmentation can
Since the system uses RGB images, it is light address lighting variability for more robust per-
sensitive and best suited for indoor use. It does formance
not support continuous gesture recognition but
instead works only on single/isolated gestures. 3.4 Human-machine interactions based on
hand gesture recog- nition using deep learning
methods[4]
3.2.3 Learnings
3.4.1 Methodologies
The approach of StaDNet can thus adapt for
the presentation control system with pose-driven The paper uses CNNs for extracting key ges- ture
spatial attention and architecture CNN-LSTM features and LSTMs to capture temporal patterns,
with TensorFlow Lite running on the Raspberry improving gesture accuracy. Medi- aPipe detects
Pi. In this way, gesture recognition may be en- hand landmarks (e.g., fingertips) in real-time
abled in real time, with a reliability improve- video, enabling gesture-based con- trol. Data on
ment of variable lighting conditions using intu- hand movements is mapped to specific actions,
itive gestures to support the control of slides in a and these mappings train CNN and LSTM models
presentation. for tasks like cursor control.

3.4.2 Limitations
3.3 Exploring The Techniques and
Challenges exist in achieving big labeled datasets
Applications of Hand Ges- ture Recognition for
for training while, at the same time, there are real-
Human- Computer Interaction [3]
time processing challenges on de- vices like the
3.3.1 Methodologies Raspberry Pi. Gesture recognition performance
may also be sensitive to environ- mental factors,
The paper utilizes OpenCV and Mediapipe for such as lighting and background, which can impact
real-time video capture and hand tracking. Me- the accuracy.

© 2024, IJSREM | www.ijsrem.com DOI: 10.55041/IJSREM39636 | Page 3

 International Journal of Scientific Research in Engineering and Management (IJSREM)
Volume: 08 Issue: 12 | Dec - 2024 SJIF Rating: 8.448 ISSN: 2582-3930

3.6 Gesture-based Human-

3.4.3 Learnings Computer Interaction usingWearable Devices[6]
The CNN-LSTM framework and MediaPipe- 3.6.1 Methodologies
based tracking can be utilized for enhanced ges-
ture accuracy of slide control on a Raspberry The paper discusses methods for collecting and
Pi. Obvious guidelines gained in this paper form processing gesture data with wearable de- vices,
guidelines for how to adapt to lighting variabil- including feature extraction using Prin- cipal
ity, and how the system might operate reliably: a Component Analysis and Fourier Trans- forms. It
good foundation on which to build a successful explores machine learning models like CNNs,
presentation control. RNNs, and Hidden Markov Models for gesture
recognition, along with real-time pro- cessing
3.5 Carry Out Computer Tasks with techniques such as parallel processing and sensor
Gesture using Image Pre- Processing and data fusion to enhance accuracy and reduce latency
TensorFlow Framework [5] on low-power devices.
3.5.1 Methodologies
Limitations
The paper uses Image Segmentation and it is able One of the main disadvantages is that wear- ables
to separate the hand against the back- ground differ significantly from one another in terms of
using techniques such as background subtraction effectiveness and dependency on hard- ware; more
and binary thresholding, will be suitable for the than likely the environment- light- ing,
accurate detection of the hand in dynamic impediments- impacts the accuracy of ges- ture
scenarios. Besides that, the paper proposes detection although low processing could possibly
Gesture Dataset Creation by labeling images to cause latency.
train it on presentation control ges- tures.
Furthermore, the CNN Model Training method 3.6.2 Learnings
utilizing TensorFlow can be adapted for gesture
classification, and thus real-time Gesture It is possible to control presentations using ges- ture
Prediction toward controlling presentations. recognition models such as CNNs or RNNs for
Raspberry Pi. Using real-time processing
3.5.2 Limitations techniques capable of improving responsiveness,
such as parallel processing
Some of the key restrictions are as follows: it
must be stable in the background condition, and
also sensitive to the illumination conditions af-
4 Methadology
fecting proper recognition of gestures. It is chal- The literature review methodology involved
lenging on hardware like the Raspberry Pi to identifying and synthesizing research articles,
require a GPU for smooth operation. Also on the technical reports, and case studies on gesture
list are false positives during highly dynamic recognition for interactive presentation control, deep
environments. learning models, and edge computing with
Raspberry Pi. Sources were selected from Google
3.5.3 Learnings Scholar, IEEE Xplore, and ScienceDi- rect,
focusing on studies from 2017 to 2024 us- ing key
Image segmentation-based hand detection builds
search terms like "gesture recognition,"
a presentation control-specific gesture dataset,
"Raspberry Pi," and "deep learning for hand
and it can also deploy CNNs with TensorFlow
gestures." A thematic analysis grouped studies
Lite to the edge for final inference. Thus,
into core areas such as gesture recognition frame-
increasing reliability under varying con- ditions
works, CNNs and RNNs, real-time edge com-
can be achieved using techniques such as the use
puting, and applications in interactive presenta-
of HSV color space to handle lighting as well as
tions. The review aimed to assess advancements,
background instability.
challenges, and potential for future research in
improving system efficiency and accessibility.

© 2024, IJSREM | www.ijsrem.com DOI: 10.55041/IJSREM39636 | Page 4

 International Journal of Scientific Research in Engineering and Management (IJSREM)
Volume: 08 Issue: 12 | Dec - 2024 SJIF Rating: 8.448 ISSN: 2582-3930

5 Implications and Consider- ations with a cost-effective tool. The work has very
for Future Work important implications that can be exploited to
transform technology which is more widely ac-
• Challenging Theoretical Frameworks and cessible to all sorts of users, therefore improv- ing
Assumptions: The findings may cast doubt on the systems in human-computer interaction sys- tems.
viability of utilising edge com- puting and deep
learning for real-time ges- ture detection on
devices with limited re- sources, such as
Raspberry Pi.
References
• Suggestions for Subsequent Research:
Future research could concentrate on inves- [1] Ali A, Abed SA, Rahman O. Python-based
tigating hybrid models (such as CNNs and RNNs) Raspberry Pi for Hand Gesture Recognition.
for improved dynamic gesture recog- nition and International Journal of Computer Applica- tions.
optimising models to lower la- tency and 2017;173(4):18–24.
computational load.
[2] Osama M, Sofiane R, Andrea C. A Deep
• Implications for Practitioners: Practi- Learning Framework for Recognizing Both Static
tioners could use these findings to create more and Dynamic Gestures. Sensors. 2021;21(6):2227.
efficient and accessible gesture-based control [3] Anonymous. Exploring The Techniques and
systems, particularly for hands-free Applications of Hand Gesture Recognition for
environments. Human-Computer Interaction. Indian Scientific
Journal Of Research In Engineer- ing And
• Influence on Programs, Methods, and
Management. 2023;07(05).
Interventions: This study could lead to more
inclusive and user-friendly technolo- gies for [4] Zholshiyeva L, Manbetova Z, Kaibassova D,
individuals with disabilities or for applications in Kassymova A, Tashenova Z, Baizhumanov S, et al.
interactive settings. Human-machine interactions based on hand gesture
recognition using deep learning methods.
• Contribution to Solving Problems: It International Journal of Electrical and Computer
may help solve accessibility issues by devel- oping Engineering. 2024.
technology that doesn’t rely on tradi- tional input
devices, improving inclusivity. [5] Magar P, Singh S, Mishra S, Gaikwad P.
Carry Out Computer Tasks with Gesture us- ing
• Improvements or Changes Needed: Image Pre-Processing and TensorFlow
Improvements should focus on enhancing gesture Framework. International Journal of Engi- neering
recognition robustness, addressing environmental Research and. 2020;9(09).
challenges, and optimizing real-time processing on
devices like Rasp- berry Pi. [6] Bansal JL. Gesture-based Human-Computer
Interaction using Wearable Devices. Interna- tional
Journal for Research Publication and Seminar.
6 Conclusions
2023.
This work applies deep learning and edge com-
puting on a Raspberry Pi to exhibit the signif-
icance of gesture detection in interactive pre-
sentation control. Currently, the study focuses on
the requirement of hands-free control sys- tems
applied in many areas, including assistive
technology and public speaking. This disser-
tation aims to look into the implementation of
deep learning models, namely CNNs and RNNs, on
edge devices so as to provide real-time ges- ture
recognition facilities for interactive presen- tations.
The chosen approach employs deep learning
frameworks and picture pre-processing utilizing
Raspberry Pi to achieve real-time performance

© 2024, IJSREM | www.ijsrem.com DOI: 10.55041/IJSREM39636 | Page 5