Modified generative adversarial network and Pseudo- Zernike matrix features extraction for human-computer interactive gesture recognition
Gesture recognition aims to understand the dynamic gestures of human body, and is one of the most important interaction methods in the field of human-computer interaction. In order to improve the current situation that the existing gesture recognition algorithms need a lot of training data, this pap...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Tamkang University Press
2025-06-01
|
| Series: | Journal of Applied Science and Engineering |
| Subjects: | |
| Online Access: | http://jase.tku.edu.tw/articles/jase-202602-29-02-0018 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Gesture recognition aims to understand the dynamic gestures of human body, and is one of the most important interaction methods in the field of human-computer interaction. In order to improve the current situation that the existing gesture recognition algorithms need a lot of training data, this paper proposes a modified generative adversarial network (GAN) and Pseudo-Zernike matrix features extraction method to solve the problems of low recognition accuracy and complex recognition process. Firstly, the improved InceptionV2 and InceptionV2-trans structures are added to the encoder and decoder respectively to enhance the feature reduction capability of the model. Secondly, conditional batch normalization is performed in each component network to improve over-fitting, and Mish activation function is used instead of ReLU function to improve network performance.
Meanwhile, we use the Pseudo-Zernike matrix to extract local features to enhance the feature extraction ability. Experiments show that the proposed method can obtain higher classification accuracy with fewer samples, and the convergence time is short, which verifies the reliability of the proposed method. |
|---|---|
| ISSN: | 2708-9967 2708-9975 |