Scholarlink

<p>Existing methods for 3D human pose estimation often suffer from localization errors primarily caused by inadequate modeling of bone length dependencies, self-occlusions, depth ambiguities, and pose diversity. To address this issue, we propose a novel refinement method that leverages anthropometric constraints and synthetic localization errors. Our approach comprises three modules: a synthetic error-guided pose refiner that fine-tunes coarse 3D poses, an anthropometric stature regressor that predicts the closest anthropometric stature, and an anthropometric pose refiner that further minimizes localization errors. Evaluated on the Human3.6M and MPI-INF-3DHP datasets, our method outperforms state-of-the-art 3D human pose estimation techniques, reducing localization errors by up to 2.5 mm and 1.3 mm, respectively. The proposed refine- ment method not only enhances the accuracy of 3D pose estimation but also demonstrates improved generalization capabilities. The code and data are publicly available at: <a href="https://github.com/alimanjotho/poseperfect">https://github.com/alimanjotho/poseperfect</a>.</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

PosePerfect: refining 3D human pose estimation using anthropometric constraints and synthetic localization errors

  • 期刊论文

  • 发表日期：

    2025年12月10日

查看全文
  • Anam Memon,
  • Qasim Arain,
  • Ali Asghar Manjotho,
  • Meshari Huwaytim Alanazi,
  • Mueen Uddin,
  • Mohammad Shorfuzzaman

摘要

Existing methods for 3D human pose estimation often suffer from localization errors primarily caused by inadequate modeling of bone length dependencies, self-occlusions, depth ambiguities, and pose diversity. To address this issue, we propose a novel refinement method that leverages anthropometric constraints and synthetic localization errors. Our approach comprises three modules: a synthetic error-guided pose refiner that fine-tunes coarse 3D poses, an anthropometric stature regressor that predicts the closest anthropometric stature, and an anthropometric pose refiner that further minimizes localization errors. Evaluated on the Human3.6M and MPI-INF-3DHP datasets, our method outperforms state-of-the-art 3D human pose estimation techniques, reducing localization errors by up to 2.5 mm and 1.3 mm, respectively. The proposed refine- ment method not only enhances the accuracy of 3D pose estimation but also demonstrates improved generalization capabilities. The code and data are publicly available at: https://github.com/alimanjotho/poseperfect.