{
  "paper_id": "6e0ecc69c6052c47251e9f50df78b82b",
  "paper_title": "A Layered Approach to People Detection in 3D Range Data",
  "paper_authors": [
    "Kai O. Arras",
    "Luciano Spinello",
    "Rudolph Triebel",
    "Roland Siegwart"
  ],
  "publication_date": "",
  "publication_venue": "Proceedings of the AAAI Conference on Artificial Intelligence",
  "abstract": "People tracking is a key technology for autonomous systems, intelligent cars and social robots operating in populated environments. What makes the task difficult is that the appearance of humans in range data can change drastically as a function of body pose, distance to the sensor, self-occlusion and occlusion by other objects. In this paper we propose a novel approach to pedestrian detection in 3D range data based on supervised learning techniques to create a bank of classifiers for different height levels of the human body. In particular, our approach applies AdaBoost to train a strong classifier from geometrical and statistical features of groups of neighboring points at the same height. In a second step, the AdaBoost classifiers mutually enforce their evidence across different heights by voting into a continuous space. Pedestrians are finally found efficiently by mean-shift search for local maxima in the voting space. Experimental results carried out with 3D laser range data illustrate the robustness and efficiency of our approach even in cluttered urban environments. The learned people detector reaches a classification rate up to 96% from a single 3D scan.",
  "keywords": [
    "people detection",
    "3D range data",
    "AdaBoost",
    "voting space",
    "mean shift"
  ],
  "paper_institutions": [
    "University of Freiburg, Germany",
    "ETH Zurich, Switzerland"
  ],
  "output_language": "zh",
  "provider": "doubao",
  "model": "ark-code-latest",
  "prompt_template": "simple",
  "extracted_at": "2026-08-16T19:47:34.402059Z",
  "template_tag": "simple",
  "summary": "### 摘要\n本文提出了一种基于分层方法的3D距离数据行人检测方法。针对行人在距离数据中的外观随姿态、距离、自遮挡和物体遮挡而剧烈变化的问题，方法将人体按高度划分为多个部分，对每个部分使用AdaBoost训练独立的强分类器，基于几何和统计特征对同一高度的邻近点组（即段）进行分类。随后，各个分类器通过向连续空间投票来相互加强证据，利用均值漂移搜索局部极大值，高效定位行人。所用的主要机制是AdaBoost强分类器 H(f)=sign(Σ α_t h_t(f))，其中h_t是决策树桩。实验在苏黎世联邦理工的两个户外数据集（Polyterrasse和Tannenstrasse）上进行，使用Velodyne HDL-64E激光扫描仪，训练数据包含203个行人，划分为9个高度层。在10米范围内，Polyterrasse数据集的等错误率（EER）达到96%，Tannenstrasse数据集达到95%，随着距离增加性能下降，但即使在20米范围内仍保持63-65%的EER。该方法无需地面平面假设，对部分遮挡具有鲁棒性，但未结合时间信息。未来工作将结合跟踪以整合时间检测结果。\n\n### 关键要点\n- 提出了一种分层的人体检测方法，将人体按高度划分为9个部分，每个部分独立训练AdaBoost分类器。\n- 每个段的特征包括宽度、点数、圆形度、线性度等17个几何和统计特征。\n- 使用跳跃距离聚类（JDC）对每个扫描层进行分割，阈值θ_d=40cm。\n- 训练使用203个行人样本，包含站和行走状态，来自Polyterrasse数据集。\n- 每个部分分类器使用决策树桩作为弱分类器，仅学习20个决策树桩以避免过拟合。\n- 由于正负样本不平衡，采用自适应的初始权重向量，负样本权重减半。\n- 通过将分类结果通过sigmoid转换为似然，再用于投票。\n- 投票模型中，每个段根据其所属部分对行人中心进行投票，投票经凝聚聚类压缩。\n- 在检测时，使用均值漂移在连续投票空间中寻找局部极大值作为行人中心。\n- 在Polyterrasse数据集上，10米范围内的EER为96%，15米为71%，20米为65%；在Tannenstrasse数据集上分别为95%、76%、63%。\n- 该算法无需地面平面假设，对遮挡具有鲁棒性，能从单一3D扫描中检测行人。\n- 该算法在未优化的C++实现中，平均检测频率约为1Hz，适用于10米范围内的点云。"
}