{
  "paper_id": "ebddd9e83413c0b353d9cbf69af87b5a",
  "paper_title": "Deep Learning for 3D Human Pose Estimation and Mesh Recovery: A Survey",
  "paper_authors": [
    "Yang Liu",
    "Changzhen Qiu",
    "Zhiyong Zhang"
  ],
  "publication_date": "",
  "publication_venue": "Neurocomputing",
  "abstract": "3D human pose estimation and mesh recovery have attracted widespread research interest in many areas, such as computer vision, autonomous driving, and robotics. Deep learning on 3D human pose estimation and mesh recovery has recently thrived, with numerous methods proposed to address different problems in this area. In this paper, to stimulate future research, we present a comprehensive review of recent progress over the past five years in deep learning methods for this area by delving into over 200 references. To the best of our knowledge, this survey is arguably the first to comprehensively cover deep learning methods for 3D human pose estimation, including both single-person and multi-person approaches, as well as human mesh recovery, encompassing methods based on explicit models and implicit representations. We also present comparative results on several publicly available datasets, together with insightful observations and inspiring future research directions. A regularly updated project page can be found at <https://github.com/liuyangme/SOTA-3DHPE-HMR>.",
  "keywords": [
    "Human pose estimation",
    "3D human pose",
    "human mesh recovery",
    "human reconstruction",
    "deep learning",
    "literature survey"
  ],
  "paper_institutions": [
    "School of Electronics and Communication Engineering, Sun Yat-sen University, Shenzhen, Guangdong, China"
  ],
  "output_language": "zh",
  "provider": "doubao",
  "model": "ark-code-latest",
  "prompt_template": "simple",
  "extracted_at": "2026-08-16T19:47:15.064648Z",
  "template_tag": "simple",
  "summary": "### 摘要\n\n本文针对三维人体姿态估计和人体网格恢复的问题，综述了近年来基于深度学习的方法。该领域涉及计算机视觉、自动驾驶、机器人等多个应用场景，但面临多人物、遮挡、深度模糊等挑战。现有综述或只关注姿态估计，或只关注网格恢复，且较少涵盖隐式表示方法。为此，本文首次系统性地综述了包括单人/多人姿态估计和基于显式/隐式模型的人体网格恢复在内的深度学习方法，涉及超过200篇参考文献，并提供了方法论分类和多个公开数据集上的比较。常用的评估指标包括MPJPE、MPJAE、MPJLE和MPVPE等，其中MPJPE定义为预测关节与真实关节坐标的L2距离的平均值：MPJPE = (1/N)∑‖p_j - p_j^*‖_2。这些指标在Human3.6M、3DPW等数据集上验证了最新方法的性能。未来研究方向包括提升推理速度、处理拥挤遮挡、开发大模型、更精细的重建、环境交互重建以及可控性和可动画性等。实际应用包括运动重定向、虚拟化身、动作识别、安防监控、SLAM、自动驾驶和人机交互，但受限于传感器类型、标注数据成本和多人物遮挡等约束。\n\n### 关键要点\n\n- 该综述覆盖2019-2023年间主流会议和期刊上的3D人体姿态估计与网格恢复方法，是首个同时涵盖单人和多人姿态估计以及显式和隐式网格恢复方法的综述。\n- 姿态估计分为直接估计和2D到3D提升方法，后者通过先检测2D关键点再映射到3D空间，可降低深度模糊的影响。\n- 多人姿态估计包括自顶向下和自底向上方法，自顶向下依赖人体检测，自底向上需解决关键点关联问题，单阶段端到端方法成为趋势。\n- 网格恢复分为基于模板（如SMPL）和模板自由方法，模板方法鲁棒但细节有限，模板自由方法灵活但需要解决稳定性问题。\n- 隐式表示如PIFu、NeRF和3D高斯溅射可重建带衣物的精细表面，但依赖精确姿态估计或需要初始点云。\n- 数据集方面，Human3.6M提供360万帧多视角标注，3DPW提供野外多人物视频，SURREAL为合成数据集，MPI-INF-3DHP提供户外场景等，用于训练和评估。\n- 评估指标包括MPJPE（关节位置误差）、MPJAE（关节角度误差）、MPJLE（关节定位误差）和MPVPE（顶点位置误差），其中MPJPE是最常用的。\n- 在Human3.6M数据集上，FTCM方法取得MPJPE 28.1mm，表现领先；在3DPW数据集上，ReFit方法取得PA-MPJPE 41.0mm。\n- 针对数据缺乏，采用无监督、自监督、弱监督、元学习和数据增强等策略减少对3D标注的依赖。\n- 应用场景包括运动重定向、虚拟化身、动作识别、安防监控、SLAM、自动驾驶和人机交互，但需考虑计算效率和边缘设备部署。\n- 挑战包括速度（边缘设备）、拥挤遮挡（多人物场景）、大模型（预训练表示）、更精细重建（细节与鲁棒性平衡）、环境交互重建、可控性和可动画性。"
}