{
  "paper_id": "14f1b9fb0663f9c94ec12fdd0dad1026",
  "paper_title": "LiDAR-based human pose estimation with MotionBERT",
  "paper_authors": [
    "Zichen Zhao",
    "Chao Zhuang",
    "Jian Li",
    "Hao Sun"
  ],
  "publication_date": "",
  "publication_venue": "2024 IEEE International Conference on Mechatronics and Automation (ICMA)",
  "abstract": "Human body pose estimation has a wide range of applications in the field of artificial intelligence and is gradually applied in daily life. However, for some specific scenes and people, privacy and security need to be ensured, and devices such as cameras are no longer the optimal choice, so this paper proposes the idea of using LiDAR technology to acquire human body point cloud images, and preprocessing and segmentation to extract a pure human body point cloud model. First, the useful parts around the human body are extracted by preprocessing the point cloud image obtained from LiDAR scanning. Then, a segmentation technique is utilized to separate the human body from the surrounding objects, keeping only the human body part. Next, the excess noise is removed by filtering process to get the pure human body point cloud model. Finally, the motionBert3d model is used to estimate the pose of the preprocessed point cloud image, and the pose estimation points and the 3D pose estimation model of the human body are obtained, which lays the foundation for the subsequent human motion recognition.\n",
  "keywords": [
    "LiDAR",
    "point cloud maps",
    "image preprocessing",
    "human pose estimation",
    "motion Bert3d modeling"
  ],
  "paper_institutions": [
    "College of Artificial Intelligence, Hebei University of Technology",
    "National Rehabilitation Aids Research Center"
  ],
  "output_language": "zh",
  "provider": "doubao",
  "model": "deepseek-v4-flash",
  "prompt_template": "simple",
  "extracted_at": "2026-08-18T10:30:52.112667Z",
  "template_tag": "simple",
  "summary": "### 摘要\n\n本文针对特定场景下人体姿态估计中的隐私与安全问题，提出基于 LiDAR 点云与 MotionBERT 的 3D 人体姿态估计方法。研究背景方面，相机、RFID、WiFi、雷达等接触式/非接触式设备各有局限，而 LiDAR 具有覆盖广、不受光照影响且不采集个人图像的优势。方法上，使用 Livox Mid-70 采集点云，通过直通滤波锁定人体区域，结合 RANSAC 与最小二乘法分割座椅平面（迭代次数由 T=log(1-p)/log(1-(1-e)^s) 决定），再经统计离群点去除与半径离群点去除混合滤波得到纯净人体点云，最后用 MotionBERT（DSTformer）估计姿态。数据集在实验室复杂环境中采集，5名志愿者完成站立、坐、蹲、行走、躺卧等姿态各约100次，并改变室内家具位置以模拟真实环境；该数据集填补了 LiDAR 人体姿态识别公开数据集的空白。实验表明，MotionBERT 能准确预测头、肩、肘、手、躯干、膝、脚等关键点，与相机方法结果相当，且更好保护隐私；但侧躺姿态存在仅识别上半身、手臂点缺失以及滤波残留噪声的问题。未来工作包括改进预处理、在 MotionBERT 中引入 Coordinate Attention 空间位置信息，以及增加侧躺等极端姿态样本。该方法适用于需要隐私保护的室内人体姿态监测场景，但当前对遮挡和复杂姿态的鲁棒性仍需提升。\n\n### 关键要点\n\n- 论文使用 Livox Mid-70 3D LiDAR 采集人体点云，其视场角为 70.4°（圆形），距离精度 2 cm，角精度小于 0.1°，数据率 100000 点/秒。\n- 预处理首先采用 PassThrough 直通滤波，通过设定过滤条件去除人体周围无关区域，锁定人体所在区域。\n- 使用 RANSAC 与最小二乘法结合，从点云中分割座椅平面，迭代次数由 T=log(1-p)/log(1-(1-e)^s) 计算，从而分离人体与周围物体。\n- 混合滤波由统计离群点去除（SOR）和半径离群点去除（ROR）组成，分别基于平均距离阈值 d_l=d_h+λ·s 和邻域点数阈值去除噪声与离群点。\n- 实验数据集在河北工业大学科技园实验室采集，包含会议桌、椅子、储物柜、投影仪等复杂环境，5名志愿者完成站立、坐、蹲、行走、躺卧五种姿态，每种约100次。\n- 数据采集时改变了室内会议桌、座椅和储物柜的位置，以模拟环境变化对点云数据的影响。\n- 姿态估计采用 MotionBERT 模型，其运动编码器基于双流时空 Transformer（DSTformer），可捕捉骨骼关节点之间的长时时空关系。\n- 实验结果显示，对于站立、坐、行走、蹲、躺卧等基本姿态，MotionBERT 能准确预测头部、双肩、肘部、双手、躯干、膝盖和双脚等关键关节点。\n- 与基于相机图像的姿态估计相比，LiDAR 扫描不直接采集个人图像，在保护个人隐私方面具有显著优势，且估计结果与相机方法相当。\n- 当前主要局限是侧躺姿态下模型仅识别上半身，无法生成双臂完整姿态估计点，且侧躺点云滤波后仍残留噪声，影响估计精度。\n- 未来工作包括增强预处理以去除更多噪声、在 MotionBERT 中引入 Coordinate Attention 机制以利用空间位置信息，以及增加侧躺等极端姿态训练样本。",
  "available_templates": [
    "deep_read",
    "simple"
  ]
}