{
  "paper_id": "d3809aba54849177dcecc6810935432a",
  "paper_title": "Socially Constrained Tracking in Crowded Environments Using Shoulder Pose Estimates",
  "paper_authors": [
    "Alexander Virgona",
    "Alen Alempijevic",
    "Teresa Vidal-Calleja"
  ],
  "publication_date": "2018",
  "publication_venue": "IEEE",
  "abstract": "Detecting and tracking people is a key requirement in the development of robotic technologies intended to operate in human environments. In crowded environments such as train stations this task is particularly challenging due to the high number of targets and frequent occlusions. In this paper we present a framework for detecting and tracking humans in such crowded environments in terms of 2D pose (x, y, θ). The main contributions are a method for extracting pose from the most visible parts of the body in a crowd, the head and shoulders, and a tracker which leverages social constraints regarding people's orientation, movement and proximity to one another, to improve robustness in this challenging environment. The framework is evaluated on two datasets: one captured in a lab environment with ground truth obtained using a motion capture system, and the other captured in a busy inner city train station. Pose errors are reported against the ground truth and the tracking results are then compared with a state-of-the-art person tracking framework.",
  "keywords": [
    "person tracking",
    "social constraints",
    "shoulder pose estimation",
    "depth camera",
    "crowded environments",
    "proxemics"
  ],
  "paper_institutions": [
    "Centre for Autonomous Systems",
    "University of Technology Sydney"
  ],
  "output_language": "zh",
  "provider": "doubao",
  "model": "ark-code-latest",
  "prompt_template": "simple",
  "extracted_at": "2026-08-16T19:47:27.597200Z",
  "template_tag": "simple",
  "summary": "### 摘要\n本论文提出一个在拥挤环境中利用深度图像进行人体检测与跟踪的框架，核心是提取头部和肩部的二维姿态（位置与朝向），并通过社会约束（如个人空间）改进跟踪鲁棒性。作者将姿态提取建模为对头部和肩部椭球体的拟合，利用人体测量学数据设定物理约束；跟踪部分采用每个行人一个粒子滤波器，在预测阶段引入社交约束（删除靠近他人均值位置的粒子并重采样），在数据关联中使用两阶段最近邻方法，并在观测更新中处理朝向模糊（通过β=0.7的歧义比率）及基于肩部椭球偏心率调整观测噪声。实验使用实验室数据集（含光学动捕真值）和悉尼市中心火车站真实数据，结果表明姿态提取在水平、垂直和朝向的误差分别约为8–17 cm、3–6 cm和9–19°，朝向正确率约70–85% ；在CLEAR-MOT指标下，所提跟踪器在密集场景（Alighting、Walkthrough、Passing）的MOTA优于Linder等人[18]的跟踪器，尤其在复杂拥挤场景提升明显，但在Wandering序列性能相近。系统在高峰期30分钟序列中，人员计数与实际人数据误差通常在2人以内。未来工作包括误差建模、机器学习分类替代手动阈值、以及学习更复杂的运动模型。\n\n### 关键要点\n- 框架包含场景分割、肩部姿态提取和社会约束跟踪三个阶段，使用深度相机而非RGB以保护隐私。\n- 肩部姿态由3D位置和绕垂直轴的角度表示，通过拟合头部和肩部两个椭球提取，椭球拟合使用Li等[15]的最小二乘方法。\n- 姿态提取利用2012年美国陆军人体测量数据设定窗口大小，头部窗口21cm，肩部窗口21cm，并利用头部椭球去除颈部区域。\n- 朝向歧义通过假设头在肩前解决，但正确率约70%，跟踪算法显式处理（β=0.7）。\n- 社会约束基于亲近学（proxemics），在预测阶段删除靠近其他滤均值位置的粒子并重绘，最多重试3次。\n- 数据关联采用两阶段：先匹配有效轨迹，再匹配无效（新）轨迹，避免假阳性轨迹窃取观测。\n- 观测更新包含位置似然和朝向似然，朝向似然同时考虑正向和反向误差，并用偏心率缩放方差。\n- 额外执行肩部朝向与速度方向的伪观测更新，增强朝向估计并改善运动预测。\n- 实验室数据集包含9个序列（Wandering、Alighting、Walkthrough、Passing），使用Optitrack系统提供真值，姿态误差如表I所示。\n- 跟踪对比使用CLEAR-MOT指标（MOTP和MOTA），在Alighting、Walkthrough和Passing序列中MOTA提升显著，但MOTP略差。\n- 火车站数据集来自悉尼Town Hall车站，30分钟高峰序列中人数计数误差通常在2人以内，但存在过度估计倾向。\n- 限制包括长距离（>4m）朝向估计偏差和手动阈值设定，未来将结合误差相关因素和机器学习分类。"
}