{
  "paper_id": "6d244b4fa26ed5401044f1745dc6cf65",
  "paper_title": "Kinect-based People Detection and Tracking from Small-Footprint Ground Robots",
  "paper_authors": [
    "Armando Pesenti Gritti",
    "Oscar Tarabini",
    "Jérôme Guzzi",
    "Gianni A. Di Caro",
    "Vincenzo Caglioti",
    "Luca M. Gambardella",
    "Alessandro Giusti"
  ],
  "publication_date": "",
  "publication_venue": "IROS 2014",
  "abstract": "Small-footprint mobile ground robots, such as the popular Turtlebot and Kobuki platforms, are by necessity equipped with sensors which lie close to the ground. Reliably detecting and tracking people from this viewpoint is a challenging problem, whose solution is a key requirement for many applications involving sharing of common spaces and close human-robot interaction. We present a robust solution for cluttered indoor environments, using an inexpensive RGB-D sensor such as the Microsoft Kinect or Asus Xtion. Even in challenging scenarios with multiple people in view at once and occluding each other, our system solves the person detection problem significantly better than alternative approaches, reaching a precision, recall and F1-score of 0.85, 0.81 and 0.83, respectively. Evaluation datasets, a real-time ROS-enabled implementation and demonstration videos are provided as supplementary material.",
  "keywords": [
    "People detection",
    "People tracking",
    "RGB-D sensor",
    "Small-footprint ground robot",
    "Leg detection",
    "SVM classifier",
    "Kalman filter"
  ],
  "paper_institutions": [],
  "output_language": "zh",
  "provider": "doubao",
  "model": "ark-code-latest",
  "prompt_template": "simple",
  "extracted_at": "2026-08-16T19:46:34.854407Z",
  "template_tag": "simple",
  "summary": "### 摘要\n\n本文针对小型地面移动机器人（如Turtlebot和Kobuki）因传感器贴近地面而带来的人员检测与跟踪难题，提出了一种基于廉价RGB-D传感器（如Kinect或Asus Xtion）的鲁棒解决方案。在杂乱室内环境中，当多人同时出现且相互遮挡时，系统利用腿部作为主要可见特征，通过候选腿分割、基于HOG特征的SVM分类以及两级跟踪框架（腿部跟踪和人员跟踪）实现了显著优于现有方法的性能，在困难场景下达到精度0.85、召回率0.81和F1值0.83。系统在三个不同复杂度场景（静止、多人、移动平台）中进行了验证，并提供了实时ROS实现和补充材料。核心机制包括：先利用DBSCAN聚类提取距地面20cm内的候选脚部，再扩展至膝盖高度并计算矫正深度图像的HOG特征，由软间隔SVM（RBF核）输出腿部概率，最后通过基于最近邻和PDAF的跟踪器融合时空信息，并利用腿部对关联生成人员轨迹。系统支持实时运行（10-30 fps），并在低帧率下性能保持稳定。未来计划利用RGB数据改善数据关联和人员重识别。\n\n### 关键要点\n\n- 问题背景：小型地面机器人传感器位置低，只能看到腿部，且存在严重遮挡和杂乱物体干扰，现有算法不适用。\n- 解决方案：采用RGB-D传感器，基于腿部检测与跟踪，结合机器学习分类器和多级滤波。\n- 候选生成：使用DBSCAN聚类地面以上20cm内的点云，再扩展至膝盖高度（50cm），并通过Voronoi区域限制扩展范围。\n- 特征提取：对每个候选腿部计算矫正深度图像（40x50cm网格），提取4x5窗口的9-bin HOG特征（180维）。\n- 分类器：采用RBF核的软SVM输出腿部概率，训练集包含26000个实例（35%为真实腿部），来自15个不同环境。\n- 腿部跟踪：使用恒速模型和卡尔曼滤波，采用最近邻关联，结合PDAF处理模糊关联，并将观测转换到固定参考系。\n- 人员跟踪：利用腿部轨迹对，通过向上扩展至髋部判断配对，结合PDAF生成人员轨迹，保守策略创建新轨迹。\n- 性能指标：在S-Difficult场景下，腿部检测精度0.96、召回率0.79、F1值0.87；人员检测精度0.85、召回率0.81、F1值0.83。\n- 对比实验：与2D激光法（IAV、Supervised Learning）和3D法（HLSN）对比，系统在精度-召回率曲线上全面领先。\n- 参数效应：腿部概率阈值τ在0.5-0.9范围内F1值稳定，取τ=0.8；帧率降至10fps时性能几乎不变，但低于10fps则显著退化。\n- 实现与运行：MATLAB实现，支持ROS，处理速度10-30fps（低端i5）或达到25fps+（i7-4930MX），数据集和代码公开。\n- 局限性：未使用RGB信息，无法进行人员重识别；假设地面平坦，适用于室内环境。"
}