Scholarlink

<p>Dynamic hand gesture recognition has gained significant attention in computer vision due to its critical role in human-computer interaction, automation, and related fields. A key challenge in this domain is the substantial inter-class and intra-class variation inherent in gesture execution. To address this, we propose a robust ensemble-based model that mitigates such variability using a novel loss function termed Discriminant Distribution-Agnostic Loss (DDA Loss). Our ensemble incorporates three diverse deep learning architectures-DenseNet121, InceptionV3, and VGG16-each independently extracting feature vectors from skeleton trajectory representations of hand gestures. These features are fused and optimized using the DDA loss, which combines center loss with a distribution-agnostic constraint to enhance class separability without assumptions on data distribution. The training is performed using the Adam optimizer. The proposed model achieved an accuracy of <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(99.8\%\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>99.8</mn> <mo>%</mo> </mrow> </math></EquationSource> </InlineEquation> on the 26-Gestures dataset, <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(98.2\%\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>98.2</mn> <mo>%</mo> </mrow> </math></EquationSource> </InlineEquation> on the DHG14/28 dataset, and <InlineEquation ID="IEq3"> <EquationSource Format="TEX">\(94.8\%\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>94.8</mn> <mo>%</mo> </mrow> </math></EquationSource> </InlineEquation> on the FPHA dataset, demonstrating performance comparable to state-of-the-art methods.</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

Ensemble learning with DDA loss for managing inter and intra class variation in hand gesture recognition

  • 期刊论文

  • 发表日期：

    2026年3月11日

查看全文
  • Reena Tripathi,
  • Bindu Verma

摘要

Dynamic hand gesture recognition has gained significant attention in computer vision due to its critical role in human-computer interaction, automation, and related fields. A key challenge in this domain is the substantial inter-class and intra-class variation inherent in gesture execution. To address this, we propose a robust ensemble-based model that mitigates such variability using a novel loss function termed Discriminant Distribution-Agnostic Loss (DDA Loss). Our ensemble incorporates three diverse deep learning architectures-DenseNet121, InceptionV3, and VGG16-each independently extracting feature vectors from skeleton trajectory representations of hand gestures. These features are fused and optimized using the DDA loss, which combines center loss with a distribution-agnostic constraint to enhance class separability without assumptions on data distribution. The training is performed using the Adam optimizer. The proposed model achieved an accuracy of \(99.8\%\) 99.8 % on the 26-Gestures dataset, \(98.2\%\) 98.2 % on the DHG14/28 dataset, and \(94.8\%\) 94.8 % on the FPHA dataset, demonstrating performance comparable to state-of-the-art methods.