{
  "paper_id": "94154a638685a33b48033b6532f6400c",
  "paper_title": "Deep Learning-based Human Pose Estimation: A Survey",
  "paper_authors": [
    "Ce Zheng",
    "Wenhan Wu",
    "Chen Chen",
    "Taojiannan Yang",
    "Sijie Zhu",
    "Ju Shen",
    "Nasser Kehtarnavaz",
    "Mubarak Shah"
  ],
  "publication_date": "August 2023",
  "publication_venue": "ACM Computing Surveys",
  "abstract": "Human pose estimation aims to locate the human body parts and build human body representation (e.g., body skeleton) from input data such as images and videos. It has drawn increasing attention during the past decade and has been utilized in a wide range of applications including human-computer interaction, motion analysis, augmented reality, and virtual reality. Although the recently developed deep learning-based solutions have achieved high performance in human pose estimation, there still remain challenges due to insufficient training data, depth ambiguities, and occlusion. The goal of this survey is to provide a comprehensive review of recent deep learning-based solutions for both 2D and 3D pose estimation via a systematic analysis and comparison of these solutions based on their input data and inference procedures. More than 260 research papers since 2014 are covered in this survey. Furthermore, 2D and 3D human pose estimation datasets and evaluation metrics are included. Quantitative performance comparisons of the reviewed methods on popular datasets are summarized and discussed. Finally, the challenges involved, applications, and future research directions are concluded. A regularly updated project page is provided: <https://github.com/zczcwh/DL-HPE>.",
  "keywords": [
    "Survey of human pose estimation",
    "2D and 3D pose estimation",
    "deep learning-based pose estimation",
    "pose estimation datasets",
    "pose estimation metrics"
  ],
  "paper_institutions": [
    "University of Central Florida, USA",
    "University of North Carolina at Charlotte, USA",
    "University of Dayton, USA",
    "University of Texas at Dallas, USA"
  ],
  "output_language": "zh",
  "provider": "doubao",
  "model": "ark-code-latest",
  "prompt_template": "simple",
  "extracted_at": "2026-08-16T19:47:45.160715Z",
  "template_tag": "simple",
  "summary": "### 摘要\n本综述系统性地回顾了基于深度学习的二维和三维人体姿态估计方法，覆盖了自2014年以来超过260篇论文。该领域旨在从图像或视频中定位人体部位并构建人体表示，广泛应用于人机交互、运动分析、增强现实和虚拟现实等场景。现有方法面临训练数据不足、深度模糊和遮挡等挑战。综述按输入数据和推理过程进行分类：2D姿态估计分为单人（回归或热图）和多人（自顶向下或自底向上），3D姿态估计则分为单目RGB、多视角及其他传感器（深度、IMU、射频等）方法。本文详细分析了各类方法的框架与创新，如回归方法直接学习映射，热图方法通过高斯核预测关键点位置，2D-to-3D提升方法借助2D姿态获得3D信息，以及基于SMPL等体积模型的人体网格恢复。数据集方面，介绍了MPII、COCO、PoseTrack、Human3.6M等，并总结了常用的评估指标如PCKh、AP、MPJPE、3DPCK。通过对比结果表明，自顶向下方法通常精度更高，而自底向上方法计算效率更优；多视角设置可缓解遮挡和深度歧义。尽管取得了显著进展，模型泛化、遮挡鲁棒性、计算效率及时序一致性仍是待解决的问题。未来方向包括域适应、轻量化人体模型、场景交互建模、新评估指标、分辨率感知、对抗防御、神经架构搜索等。本文还介绍了姿态估计在动作识别、视频监控、在线教练、服装解析、动画、AR/VR和医疗健康等领域的应用。\n\n### 关键要点\n- 2D单人姿态估计包含回归方法和热图方法，热图方法通过高斯核生成关键点概率图，训练更平滑。\n- 2D多人姿态估计分为自顶向下（先检测再单人或估计）和自底向上（先检测所有关键点再分组），自顶向下通常精度更高但计算量随人体数增加。\n- 3D姿态估计从单目图像直接估计或通过2D姿态提升，2D-to-3D提升通常优于直接估计，得益于成熟2D检测器。\n- 人体网格恢复（HMR）利用SMPL等参数化模型重建三维网格，但参数众多，需权衡效率与质量。\n- 多视角3D姿态估计利用几何约束缓解遮挡，但需解决视角关联，如VoxelPose和Faster VoxelPose。\n- 其他传感器（深度、IMU、射频）可提供补充信息，但受成本或分辨率限制。\n- 常用2D数据集包括MPII、COCO、PoseTrack，3D数据集包括Human3.6M、MuPoTS-3D等，3D标注采集困难。\n- 评估指标方面，2D常用PCKh、AP、mAP，3D常用MPJPE、PA-MPJPE、3DPCK，但现有指标无法评估时序平滑性。\n- 实验对比显示，2D自顶向下方法在COCO上AP更高（如Cai等79.2），自底向上方法如PifPaf在低分辨率下表现较好。\n- 3D单视角多人在MuPoTS-3D上，集成方法如Chen等达到89.6 3DPCK，高于纯自顶向下或自底向上。\n- 多视角方法在Human3.6M上MPJPE可低至18.7mm（2021年），远优于单视角方法，证实多视图优势。\n- 现有挑战包括遮挡、罕见姿态数据不足、计算效率、模型泛化到野外场景，以及跨域适应和对抗攻击鲁棒性。"
}