{
  "paper_id": "36d07681dc68fe4d6fb0649c2781fb1d",
  "paper_title": "DR-SPAAM: A Spatial-Attention and Auto-regressive Model for Person Detection in 2D Range Data",
  "paper_authors": [
    "Dan Jia",
    "Alexander Hermans",
    "Bastian Leibe"
  ],
  "publication_date": "",
  "publication_venue": "2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)",
  "abstract": "Detecting persons using a 2D LiDAR is a challenging task due to the low information content of 2D range data. To alleviate the problem caused by the sparsity of the LiDAR points, current state-of-the-art methods fuse multiple previous scans and perform detection using the combined scans. The downside of such a backward looking fusion is that all the scans need to be aligned explicitly, and the necessary alignment operation makes the whole pipeline more expensive – often too expensive for real-world applications. In this paper, we propose a person detection network which uses an alternative strategy to combine scans obtained at different times. Our method, Distance Robust SPatial Attention and Auto-regressive Model (DR-SPAAM), follows a forward looking paradigm. It keeps the intermediate features from the backbone network as a template and recurrently updates the template when a new scan becomes available. The updated feature template is in turn used for detecting persons currently in the scene. On the DROW dataset, our method outperforms the existing state-of-the-art, while being approximately four times faster, running at 87.2 FPS on a laptop with a dedicated GPU and at 22.6 FPS on an NVIDIA Jetson AGX embedded GPU. We release our code in PyTorch and a ROS node including pre-trained models.",
  "keywords": [
    "person detection",
    "2D LiDAR",
    "temporal aggregation",
    "spatial attention",
    "auto-regressive model",
    "DR-SPAAM",
    "DROW dataset",
    "real-time processing"
  ],
  "paper_institutions": [
    "RWTH Aachen University"
  ],
  "output_language": "zh",
  "provider": "doubao",
  "model": "ark-code-latest",
  "prompt_template": "simple",
  "extracted_at": "2026-08-16T19:46:45.366233Z",
  "template_tag": "simple",
  "summary": "### 摘要\n本文针对2D LiDAR点云中人体检测的挑战，提出了一种结合空间注意力与自回归模型的检测网络DR-SPAAM。现有方法（如DROW）通过融合多帧扫描（反向融合）提升检测精度，但需要显式对齐（如里程计）导致计算开销大，难以实时运行。DR-SPAAM采用前向范式，维护一个特征模板并利用相似度空间注意力关联错位特征，再通过自回归模型（更新率α）累积历史信息，无需显式对齐且仅需保存单模板。在DROW数据集（24,012帧，分为训练/验证/测试）上，使用平均精度（AP）、峰值F1和等错误率（EER）评估，DR-SPAAM的AP0.5达到70.3%，优于DROW（T=5，+odom）的68.1%，且速度提升约4倍：在RTX 2080笔记本GPU上87.2 FPS，在Jetson AGX上22.6 FPS。核心机制是模板更新公式F̃_n^t = αF_n^t + (1-α)Σ softmax(Ω̃_n)_j F̃_j^{t-1}，其中Ω̃由当前特征与之前模板的相似度计算。消融实验表明空间注意力与自回归组件均有效。进一步，方法对采样频率变化鲁棒（步长5时AP0.3仅降2.1%），并可用于生成轨迹关联。该方法适用于救援、安防、医疗等机器人场景，但需注意2D LiDAR远距离信息稀疏导致的性能下降。\n\n### 关键要点\n- 2D LiDAR人体检测的主要困难是数据稀疏，现有方法通过多帧融合提升性能，但需显式对齐，计算开销大，难以实时。\n- DR-SPAAM采用前向时间聚合范式，仅保留一个特征模板，利用自回归模型更新，避免了对多帧的显式对齐和重复采样。\n- 空间注意力模块通过相似度（ψ(F_j^{t-1})^T·ψ(F_n^t)）加权融合邻域特征，学习关联错位点，抑制无关信息。\n- 自回归更新公式为：F̃_n^t = αF_n^t + (1−α)Σ softmax(Ω̃_n)_j F̃_j^{t-1}，α控制更新速率，模板累积历史信息。\n- 在DROW数据集上，DR-SPAAM的AP0.5达到70.3%，比DROW（T=5，+odom）高2.2%，比单扫描基线高3.7%。\n- DR-SPAAM在RTX 2080笔记本GPU上达到87.2 FPS，在Jetson AGX上达到22.6 FPS，远高于实时需求，而DROW（T=5）在Jetson上仅2.6 FPS。\n- 训练采用Adam优化器，初始学习率10⁻³指数衰减至10⁻⁶，用二元交叉熵和L1损失，批量大小8，训练40轮。\n- 通过超参数优化，最终选定窗口尺寸1.0m×1.0m、56个采样点，注意力窗口W=11，更新率α=0.5。\n- 消融实验显示，去掉空间注意力（DR-AM）性能降至66.3%，去掉自回归（DR-SPA）性能为68.0%，完整模型性能最佳，证明两个组件均关键。\n- DR-SPAAM对采样率变化鲁棒，步长5时AP0.3仅下降约2.1%，而DROW（T=5）下降超过15%，说明相似度注意力适应不同时间间隔。\n- 利用注意力相似度可关联相邻帧的点，生成轨迹，为跟踪和运动规划提供信息，但该功能仅初步展示，未来需深入研究。\n- 该方法适用于救援、安防、医疗等机器人场景，代码和ROS节点已开源，便于部署，但远距离检测可靠性仍受点云稀疏性限制。"
}