{
  "paper_id": "79acd19a4e9fd9836469874c5f0416c2",
  "paper_title": "Human Pose Estimation and Skeleton-Based Action Recognition: A Systematic Review of 2D/3D Deep Learning Approaches",
  "paper_authors": [
    "Kudratjon Zohirov",
    "Feruz Ruziboev",
    "Sardor Boykobilov",
    "Mirjakhon Temirov",
    "Kamoliddin Ablakulov",
    "Elbek Jabborov"
  ],
  "publication_date": "",
  "publication_venue": "2026 IEEE 9th World Conference on Computing and Communication Technologies (WCCCT)",
  "abstract": "This article presents a systematic review of modern 2D and 3D deep learning approaches used in the fields of Human Pose Estimation (HPE) and Skeleton-Based Action Recognition (SBAR). The research was conducted based on the SALSA methodology and the published works in leading scientific databases were analyzed. Within the framework of the review, approaches based on the detection of human joint points from image and video sequences, reconstruction of the skeletal structure, and modeling of actions in the spatial and spatiotemporal (ST) domains were compared. The article reviews the architectures of 2D and 3D HPE models, joint hiding in multiperson scenes, real-time requirements, and the possibilities of application in embedded devices. The spatial, spatiotemporal, and graphical features used in SBAR systems and their impact on computational complexity and energy efficiency are also analyzed. The performance of the models was compared based on evaluation criteria such as MPJPE (Mean Per Joint Position Error), AP (Average Precision), RMSE (Root Mean Square Error), and Pearson correlation. The results show that skeleton-based approaches are effective solutions for real-time and resource-constrained systems. However, choosing the optimal model and features requires a trade-off between accuracy, computational complexity, and energy efficiency. This paper identifies promising architectures, features, and hardware adaptation strategies for practical applications of HPE and SBAR systems in real-world environments.",
  "keywords": [
    "Human pose estimation",
    "2D pose estimation",
    "3D pose estimation",
    "skeleton-based action recognition",
    "deep learning",
    "multi-person scenes",
    "real-time systems",
    "computer vision",
    "biomechanics",
    "human–machine interface"
  ],
  "paper_institutions": [
    "Department of Software and hardware support of computer systems, Karshi state technical university, Karshi, Uzbekistan",
    "Department of Convergence of Digital Technologies, Tashkent University of Information Technologies, Tashkent, Uzbekistan",
    "Department of Information Systems and Technologies, Karshi State Technical University, Karshi, Uzbekistan"
  ],
  "output_language": "zh",
  "provider": "doubao",
  "model": "ark-code-latest",
  "prompt_template": "simple",
  "extracted_at": "2026-08-16T19:47:42.754315Z",
  "template_tag": "simple",
  "summary": "### 摘要\n本系统综述基于SALSA方法分析了2010至2026年间来自Scopus、IEEE Xplore等数据库的44篇文献，系统比较了2D与3D人体姿态估计（HPE）和基于骨架的动作识别（SBAR）的深度学习方法。综述涵盖数据采集技术（RGB、深度、红外、IMU等）、骨架拓扑（如OpenPose的18/25关节、BlazePose的33关节等）、特征选择（空间、时域、图域）以及嵌入式设备上的实现考量。通过MPJPE、RMSE等指标评估，显示2D方法计算轻量适合实时与资源受限设备，3D方法精度高但计算成本高，图域特征表达力强但复杂度高，多模态融合虽有更高精度但增加系统复杂性。综述强调了在准确率、计算复杂度和能效之间的权衡，并指出未来方向包括模型简化、量化和硬件级优化。\n\n### 关键要点\n- 基于SALSA方法，系统筛选并分析了2010至2026年间44篇来自Scopus、IEEE Xplore等数据库的HPE与SBAR相关文章。\n- 数据采集方法包括RGB相机、深度相机、红外相机、标记/无标记动作捕捉、IMU传感器和移动设备，各具优缺点。\n- 常见骨架模型关节数从16（LiDAR+IMU）到55（SMPL-X）不等，影响精度与计算负载。\n- 特征分为空间（JC、JD、JA等）、时空（JV、JA等）和图域（SE、AW、GN等），其中空间和时空特征最常用。\n- 嵌入式设备上常提取JC、(x,y)、(x,y,z)、HR、KCS、JAθ、JD、AW和PSE等特征，需平衡精度与效率。\n- 2D HPE模型（如OpenPose、MediaPipe）计算轻量，适合实时和嵌入式应用；3D HPE（如VideoPose3D）精度高但需更多算力。\n- 图神经网络（如ST-GCN）在复杂动作建模上表达力强，但计算复杂度高，限制其在嵌入式场景的应用。\n- 多传感器融合（如IMU+相机、LiDAR+IMU）可降低MPJPE或RMSE，例如SimpleNet在Totalcapture下MPJPE仅20.7mm，但增加系统复杂度。\n- 评价指标包括MPJPE、RMSE、AP和Pearson相关系数，不同设备和模型下MPJPE范围从20.7mm到107.5mm不等。\n- 在嵌入式设备（如Jetson TX2）上，TSHDC模型MPJPE为48.4mm，表明优化模型可在有限资源下实现合理精度。\n- 选择最优模型和特征需在准确率、计算复杂度和能耗之间权衡，尚无普适最优方案。\n- 未来研究方向包括简化模型、减少冗余参数、量化和硬件级优化，以支持实时和实际环境中的应用。"
}