Scholarlink

<p>Hand pose estimation plays a vital role in the domain of human–computer interaction. However, achieving accurate predictions in real-world scenarios, handling multi-sized targets, reducing computational burden, and striking a balance between speed and accuracy remain significant challenges for existing approaches. In this paper, we propose a MSIPA-HandNet method to address these issues. Moreover, the MSIPA-HandNet provides better speed support within the speed-accuracy trade-off. Our method consists of a lightweight encoder–decoder network architecture that is suitable for practical applications and distribution-aware coordinate representations to fine-tune the location information of gesture keypoints, with a primary emphasis on enhancing accuracy within the speed-accuracy trade-off. We also introduce a comprehensive speed-accuracy trade-off evaluation model to guide future hardware platform selection. Comprehensive experiments on two public datasets show that the proposed method outperforms current state-of-the-art approaches to hand pose estimation. Finally, we suggest that the computer terminal is a preferable deployment hardware platform, as it provides an optimal balance between speed and accuracy, and can support gesture-based location information for control applications.</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

Speed-accuracy trade-off in lightweight-based hand pose estimation

  • 期刊论文

  • 发表日期：

    2025年8月4日

查看全文
  • Mingyue Zhang,
  • Zhiheng Zhou

摘要

Hand pose estimation plays a vital role in the domain of human–computer interaction. However, achieving accurate predictions in real-world scenarios, handling multi-sized targets, reducing computational burden, and striking a balance between speed and accuracy remain significant challenges for existing approaches. In this paper, we propose a MSIPA-HandNet method to address these issues. Moreover, the MSIPA-HandNet provides better speed support within the speed-accuracy trade-off. Our method consists of a lightweight encoder–decoder network architecture that is suitable for practical applications and distribution-aware coordinate representations to fine-tune the location information of gesture keypoints, with a primary emphasis on enhancing accuracy within the speed-accuracy trade-off. We also introduce a comprehensive speed-accuracy trade-off evaluation model to guide future hardware platform selection. Comprehensive experiments on two public datasets show that the proposed method outperforms current state-of-the-art approaches to hand pose estimation. Finally, we suggest that the computer terminal is a preferable deployment hardware platform, as it provides an optimal balance between speed and accuracy, and can support gesture-based location information for control applications.