{
  "paper_id": "157cf099bde324dc70a9d8c5ddd7283c",
  "paper_title": "Li2Former: Omni-Dimension Aggregation Transformer for Person Detection in 2-D Range Data",
  "paper_authors": [
    "Haodong Yang",
    "Yadong Yang",
    "Chenpeng Yao",
    "Chengju Liu",
    "Qijun Chen"
  ],
  "publication_date": "",
  "publication_venue": "IEEE Transactions on Instrumentation and Measurement",
  "abstract": "The sparsity of plane point clouds remains human detection based on 2-D LiDAR a challenging task. Recent deep learning-based methods have made significant progress, but their unidimensional feature modeling hinders the flow of information across different axes (i.e., temporal, spatial, and batched). Besides, convolution operators limit the effective receptive field (ERF) of the model. To promote multidimensional interaction, we propose an omni-dimensional aggregation (ODA) module composed of a main aggregation stream and an auxiliary feature stream to assist in the temporal–spatial joint encoding as well as implicitly models the potential relationships between samples. By cascading multiple ODA blocks, the semantic gaps between different features are gradually bridged. In addition, to improve network convergence and generalization, an adaptive focus mechanism (AFM) is designed to guide the model to optimize more on difficult samples instead of simple or low-quality samples. Extensive experiments demonstrate the effectiveness of the proposed pipeline, achieving 76.4% and 81.4% AP on the DROW and JRDB benchmark datasets, respectively, surpassing the existing state-of-the-art (SOTA) results. Competitive results are also achieved in inference speed. Furthermore, we successfully validate the proposed model through real-world environments. Our code is available at <https://github.com/ai-winter/Li2Former-v1>.",
  "keywords": [
    "2-D LiDAR",
    "deep learning",
    "human detection and tracking"
  ],
  "paper_institutions": [
    "Tongji University",
    "Michigan State University",
    "Tongji Artificial Intelligence (Suzhou) Research Institute"
  ],
  "output_language": "zh",
  "provider": "doubao",
  "model": "ark-code-latest",
  "prompt_template": "simple",
  "extracted_at": "2026-08-16T19:47:38.944736Z",
  "template_tag": "simple",
  "summary": "### 摘要\n\n本文针对基于2-D LiDAR的行人检测中平面点云稀疏、现有深度学习方法仅进行单一维度特征建模（时间、空间或批次）导致信息流动受阻、以及卷积算子限制有效感受野的问题，提出了Li2Former模型。其核心是设计了全维度聚合（ODA）模块，包含主聚合流和辅助特征流，通过级联多个ODA块实现时间-空间-批次维度的联合编码与特征对齐，并利用Transformer的自注意力机制扩大感受野；同时提出基于高斯分布的适应性聚焦机制（AFM），动态调整难易样本的梯度权重，加快收敛并提升泛化能力。在DROW和JRDB两个基准数据集上进行实验，分别达到76.4%和81.4%的AP0.5，超越现有SOTA方法；同时通过消融实验验证了各模块的有效性，并在真实机器人平台上完成了部署验证。该方法在精度和速度之间取得了良好平衡，但受限于Transformer参数较大导致推理速度略低，且对高速运动行人的检测仍有挑战。\n\n### 关键要点\n\n- ODA模块采用双流训练策略，主流编码时间特征，辅助流编码批次-空间特征，并在推理时移除辅助流以避免训练-测试不一致。\n- 通过级联ODA块逐步弥合不同维度特征间的语义差距，提升特征融合质量。\n- 基于1-D卷积的骨干网络结合cutout预处理，将原始扫描转换为固定大小的感知窗口，适应不同距离的物体密度变化。\n- 时间编码器使用带位置编码的多头自注意力建模帧间依赖，批次-空间编码器无位置编码，实现样本间关系的位置不变学习。\n- 时间池化模块采用可学习的加权融合平均池化和最大池化，分别捕获局部和全局时间信息。\n- AFM通过高斯函数计算每个样本的回归损失权重，重点增强难样本的梯度，抑制简单样本和无效样本。\n- 在DROW数据集上，Li2Former-A达到72.5% AP0.3和76.4% AP0.5，比DROW^p分别提升1.4%和3.8%。\n- 在JRDB数据集上，Li2Former-A达到74.9% AP0.3和81.4% AP0.5，比DR-SPAAM^p提升约2.5%。\n- 通过Grad-CAM可视化，Li2Former能更清晰地区分前景行人与背景噪声，优于CNN基线。\n- 消融实验表明，ODA结构（主流T-Encoder+辅助BS-Encoder）效果最佳，而将批次与时间维度合并会破坏序列学习。\n- AFM参数α=1.5、δ=0.33时验证集性能最优，且AFM加速损失下降并保持更高精度。\n- 真实机器人实验使用PEPPERL FUCHS R2000 LiDAR，Li2Former在25.2%的FPS损失下显著降低误检率，但高速行人场景仍存在漏检。"
}