Hand Gesture Recognition from 2D Images by Using Convolutional Capsule Neural Networks
Yükleniyor...
Dosyalar
Tarih
2021
Yazarlar
Dergi Başlığı
Dergi ISSN
Cilt Başlığı
Yayıncı
Springer Heidelberg
Erişim Hakkı
info:eu-repo/semantics/closedAccess
Özet
Object classification and recognition are an important research area widely used in computer vision and machine learning. With the use of deep learning methods in the field of object recognition, there have been important developments in recent years. Object recognition and its sub-branches face recognition, motion recognition, and hand gesture recognition are now used effectively in devices used in daily life. Hand sign classification and recognition are an area that researchers are working on and trying to develop for human-computer interaction. In this study, a hybrid model was created by using a capsule network algorithm with a convolutional neural network for object classification. A dataset, named HG14, containing 14 different hand gestures was created. To measure the success of the proposed model in object recognition, training was carried out on HG14, FashionMnist, and Cifar-10 datasets. Also, VGG16, ResNet50, DenseNet, and CapsNet models were used to classify the images in HG14, FashionMnist, and Cifar-10 datasets. The results of the training were compared and evaluated. The proposed hybrid model achieved the highest accuracy rates with 90% in the HG14 dataset, 93.88% in the FashionMnist dataset, and 81.42% in the Cifar-10 dataset. The proposed model was found to be successful in all studies compared to other models.
Açıklama
Anahtar Kelimeler
Deep learning, Hand gesture recognition, CNN, Capsule network, Human-computer interaction, Object classification, Dataset
Kaynak
Arabian Journal For Science And Engineering
WoS Q Değeri
Q2
Scopus Q Değeri
Q1
