Page 456 - Kaleidoscope Academic Conference Proceedings 2024

2024 ITU Kaleidoscope Academic Conference

A convolution neural network is used in this gesture dataset using the spatial fuzzy matching
study.[11] to recognize hand gestures using data (SFM) approach [18-30] worked on a machine
from the Kinect sensor. It uses 5 people and 8 learning model to categorize Twitter posts into
different types of gestures, and its accuracy is positive, negative, and neutral categories.
98.52%.The proposed hand gesture recognition Implemented on “twitter-airline-sentiment”
system by.[12] was used to recognize the digits dataset by using Random Forest (RF) Decision
0 through 9. The effectiveness of the two Tree (DT), Naive Bayes (NB), and K-nearest
strategies was compared by the authors in this neighbors(KNN) and achieved 83%
research. The convolutional Neural Network accuracy.[31]introduce a multiscale deep
(CNN) approach was used under changeable learning model for unconstrained hand detection
conditions, such as picture rotation and scaling in still images. Deep learning models, and deep
with constant background, before the contour- convolutional neural networks(CNNs) on
SVM-based method. Three datasets of the comprehensive datasets collected from several
highest caliber, SLD, ASL, and ASL-FS, were different public image resources. And achieved
used to test the proposed methodologies. 81.25% accuracy.[32] are discussed myoelectric
Authors discovered that the contour and draw control scheme for hand prosthesis leveraging
convex hull in the contour-based approach. HD-EMG and deep learning and implemented
SVM was used for classification, and gestures in a fully embedded adaptive gesture
were identified based on the length and angle of recognition system featuring the complete chain
the convex hull. CNN relied on a method that from bio-electric sensing through deep learning
divided the digit data into five convolutional model training to real-time inference and
layers. To accomplish hand gesture operation and achieved state-of-the-art offline
identification, authors used more than 10,000 classification results with 98.2% accuracy.
digital photos from the database for the digits 1
to 10. The original data for this database was
gathered from a Creative Senz3D camera with a
320x240 resolution. The proposed technique,
according to the authors, obtained an accuracy
close to 69% for contour-SVM and 98.31% for
the CNN-based approach.[13] introduce CNN in
Sign language and find good accuracy in
comparison to other machine learning methods
currently in use.M Islam and others [14] the
impact of data augmentation in deep learning
was explored and analyzed by the writers. In
this study, the author used a self-contracted
dataset using CNN for classification, achieving
an accuracy of 98.12% for CNN with
augmented reality and 92.87% for CNN
without. CNNs were employed for classification
in [15] study of static hand gestures. The author
used two picture bases comprising 24
movements, certain segmentation techniques,
and the CNNs to get a classification accuracy of
96.83%. A deep CNN feature-based static hand
motion detection system was proposed by[16].
Deep features are retrieved using fully
connected layers of AlexNet) and the redundant Table 1. Chronological summary of various
features are subsequently minimized using PCA. techniques in the domain
SVM was then used as a classifier to categorize
the poses of hand motions. Using a dataset of 36 In this section, we present an efficient and
gesture poses, the system's performance was effective method for hand gesture recognition
assessed. The results showed an average and Model using data augmentation for
accuracy of 87.83%.The classification of hand providing better results.
gestures based on inaudible sound using
convolutional neural networks was explored
by.[17] The author used both the CNN-based
and STFT approaches, and she was able to
identify 8 different hand gestures with 87.75%
accuracy. The author was able to create a fused

– 412 –