{
  "paper_id": "8e55efe2ac55831db6d4bda72f158760",
  "paper_title": "LiDARCap: Long-range Marker-less 3D Human Motion Capture with LiDAR Point Clouds",
  "paper_authors": [
    "Jialian Li",
    "Jingyi Zhang",
    "Zhiyong Wang",
    "Siqi Shen",
    "Chenglu Wen",
    "Yuexin Ma",
    "Lan Xu",
    "Jingyi Yu",
    "Cheng Wang"
  ],
  "publication_date": "",
  "publication_venue": "2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)",
  "abstract": "Existing motion capture datasets are largely short-range and cannot yet fit the need of long-range applications. We propose LiDAR Human26M, a new human motion capture dataset captured by LiDAR at a much longer range to overcome this limitation. Our dataset also includes the ground truth human motions acquired by the IMU system and the synchronous RGB images. We further present a strong baseline method, LiDAR Cap, for LiDAR point cloud human motion capture. Specifically, we first utilize PointNet++ to encode features of points and then employ the inverse kinematics solver and SMPL optimizer to regress the pose through aggregating the temporally encoded features hierarchically. Quantitative and qualitative experiments show that our method outperforms the techniques based only on RGB images. Ablation experiments demonstrate that our dataset is challenging and worthy of further research. Finally, the experiments on the KITTI Dataset and the Waymo Open Dataset show that our method can be generalized to different LiDAR sensor settings.",
  "keywords": [
    "LiDAR",
    "human motion capture",
    "long-range",
    "point cloud",
    "dataset",
    "markerless",
    "SMPL",
    "inverse kinematics"
  ],
  "paper_institutions": [
    "Fujian Key Laboratory of Sensing and Computing for Smart Cities, Xiamen University",
    "Shanghai Engineering Research Center of Intelligent Vision and Imaging, ShanghaiTech University"
  ],
  "output_language": "zh",
  "provider": "doubao",
  "model": "ark-code-latest",
  "prompt_template": "simple",
  "extracted_at": "2026-08-16T19:46:53.477063Z",
  "template_tag": "simple",
  "summary": "### 摘要\n本文针对现有动作捕捉数据集多为短距离、难以满足远距离应用需求的问题，提出首个基于LiDAR的远距离无标记人体运动捕捉数据集LiDAR Human26M，该数据集包含同步LiDAR点云、RGB图像和由专业IMU设备采集的3D人体运动真值，覆盖20种日常动作、13名表演者、184,048帧，捕获距离从12米到28米。基于该数据集，作者提出LiDARCap基线方法，利用PointNet++提取点云全局特征，通过双向GRU进行时序编码，再经ST-GCN作为逆运动学求解器回归关节旋转（6D表示），最后通过SMPL优化器对姿态参数进行几何约束，总损失为关节位置损失、姿态损失和SMPL关节位置损失之和。实验表明，在长距离场景下，LiDARCap在MPJPE、PA-MPJPE、PCK、Accel和PVE等指标上显著优于基于图像的HMR和VIBE方法，消融实验验证了各模块的有效性，并在KITTI和Waymo数据集上验证了跨传感器泛化能力。未来可探索更复杂场景、多人和遮挡情况，以及应对点云稀疏性。其核心机制是通过级联的时序编码、逆运动学求解和SMPL优化，利用LiDAR的深度信息在远距离下获得鲁棒的人体姿态。\n### 关键要点\n- 提出LiDAR Human26M数据集，是首个专为远距离（12-28米）设计的LiDAR人体运动捕捉基准，包含184,048帧、约2600万个有效点，覆盖20种日常动作和13名表演者。\n- 数据集包含三种模态：LiDAR点云、RGB图像和IMU采集的精确3D人体运动真值，可用于多模态行为分析。\n- 数据在两类场景采集（庭院和两楼之间的开放空间），距离分布集中在17-19米（占47.2%），最远可达28米。\n- 远距离导致点云稀疏，人体点数在30到450之间变化，距离增加时会出现整身稀疏和局部缺失，如27米时手臂缺失。\n- LiDARCap采用三阶段流水线：PointNet++时空编码（结合双向GRU）、ST-GCN逆运动学求解器、SMPL优化器，端到端训练。\n- 损失函数包括关节位置损失L_J、姿态损失L_Theta和SMPL关节位置损失L_J_SMPL，三者加权联合优化。\n- 与图像方法HMR和VIBE相比，LiDARCap在MPJPE（79.31mm vs 224.86/154.61mm）和PA-MPJPE（66.72 vs 130.71/108.19）上大幅领先，PVE也显著降低。\n- 消融实验显示，用P4Transformer替换PointNet++性能相近（MPJPE 79.52），但Attention和Voting操作反而使性能下降，而ST-GCN优于GRU。\n- 增加SMPL优化器损失（L_J_SMPL）使MPJPE从87.13降至79.31，表明几何后验约束对姿态回归的必要性。\n- 距离分析表明，在28米最远距离时性能有所下降，但LiDAR方法仍优于VIBE，因为点云中人体与背景可清晰分离。\n- 在KITTI和Waymo数据集上的定性实验验证了方法对不同LiDAR传感器的泛化能力，能正确估计脚步和全局方向。\n- 局限：数据集场景平坦开放、无遮挡，缺少形状参数beta和多人体交互，方法对点云密度变化鲁棒性不足。"
}