Scholarlink

<p>Hand gesture recognition provides a robust and intuitive interface for non-verbal human–computer interaction. Despite significant advancements, existing methods often suffer from overfitting and poor generalization to new data. This is mainly due to dependence on static frames or inadequate frame selection techniques. This paper addresses these limitations by proposing a novel gesture recognition framework integrating temporal shift module (TSM) and spatio-temporal pyramid pooling (STPP) into a ResNet-50 backbone. The TSM enhances temporal dynamics, whereas STPP captures multi-scale spatio-temporal features. In addition, an entropy-based frame selection method is applied to ensure the selection of informative frames, improving the model’s ability to capture gesture dynamics. The proposed method achieves an accuracy of 85.16%, outperforming existing techniques and addressing the limitations of prior models.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing hand gesture recognition using entropy-based deep neural architecture

  • 期刊论文

  • 发表日期：

    2025年2月27日

查看全文
  • Adyasha Jena,
  • Sougatamoy Biswas,
  • Anup Nandy

摘要

Hand gesture recognition provides a robust and intuitive interface for non-verbal human–computer interaction. Despite significant advancements, existing methods often suffer from overfitting and poor generalization to new data. This is mainly due to dependence on static frames or inadequate frame selection techniques. This paper addresses these limitations by proposing a novel gesture recognition framework integrating temporal shift module (TSM) and spatio-temporal pyramid pooling (STPP) into a ResNet-50 backbone. The TSM enhances temporal dynamics, whereas STPP captures multi-scale spatio-temporal features. In addition, an entropy-based frame selection method is applied to ensure the selection of informative frames, improving the model’s ability to capture gesture dynamics. The proposed method achieves an accuracy of 85.16%, outperforming existing techniques and addressing the limitations of prior models.