{
  "paper_id": "411afd378cb9601846ca15b42fee2093",
  "paper_title": "A fast and robust 3D LiDAR-based loop closure detection method using BEV images",
  "paper_authors": [
    "Chizhou Zhang",
    "Mingsong Chen",
    "Guanqiang Wang",
    "Yongcheng Lin",
    "Zehao Li"
  ],
  "publication_date": "2026-02-12",
  "publication_venue": "International Journal of Intelligent Robotics and Applications",
  "abstract": "In this paper, a novel method for loop closure detection (LCD) is presented, which utilizes multi-layer occupied bird's eye view (BEV) images. Initially, the original point clouds undergo preprocessing and are then projected into multi-layer occupied BEV images. Subsequently, to expedite the extraction of features from the BEV images, we employ a lightweight U-shaped network incorporating the Atrous Spatial Convolution Pooling Pyramid (ASPP) module. Lastly, a similarity measurement is devised, leveraging the rotation-invariance characteristic of the descriptor. The primary innovations center around the design of the lightweight U-shaped network, successfully achieved through the use of depth-wise separable convolution (DW-conv) and larger separable convolution kernels during the down-sampling process. Additionally, the incorporation of the ASPP module before the up-sampling process enriches the feature information derived from point clouds. Experimental comparisons conducted on the NCLT and MulRan datasets highlight significant improvements in both precision and efficiency achieved by the proposed method. Notably, the inference time is reduced by more than 20% when compared to the advanced DiSCO method. Furthermore, our LCD method has been seamlessly integrated into a comprehensive LiDAR-based simultaneous localization and mapping (SLAM) system, providing validation of its effectiveness.\n",
  "keywords": [
    "Loop closure detection",
    "Bird's eye view",
    "Lightweight",
    "Atrous spatial convolution pooling pyramid"
  ],
  "paper_institutions": [
    "School of Artificial Intelligence, Hubei University of Automotive Technology",
    "School of Mechanical and Electrical Engineering, Central South University",
    "Light Alloy Research Institute, Central South University",
    "School of Mechanical and Electrical Engineering, Guilin University of Electronic Technology"
  ],
  "output_language": "zh",
  "provider": "doubao",
  "model": "deepseek-v4-flash",
  "prompt_template": "simple",
  "extracted_at": "2026-08-18T10:30:56.397088Z",
  "template_tag": "simple",
  "summary": "### 摘要\n\n本文针对基于3D LiDAR的闭环检测（LCD）中描述子判别力不足、偏航角估计缺失以及相似度计算对点云旋转敏感等问题，提出一种基于多层占用鸟瞰图（BEV）的快速鲁棒LCD方法Fast-DiSCO。相关工作表明，传统手工特征和视觉方法受视角、光照等影响，而现有深度学习方法存在参数多、推理慢等局限。该方法将原始点云预处理后投影为多层占用BEV图像，利用轻量级U形网络（编码器采用深度可分离卷积和5×5大卷积核，上采样前引入ASPP模块）提取旋转不变描述子，并通过傅里叶变换的幅值不变性实现旋转不变性，核心机制可表示为 S = |F_f ∘ F_c ∘ F_p|，同时用相位相关估计相对偏航角，最后用余弦距离进行相似度测量。在NCLT和MulRan数据集上训练与测试（BEV分辨率40×120×20，Adam优化器，四元组损失），实验表明方法在Recall@1%、Recall@1和偏航角估计误差上优于DiSCO、ScanContext、PointNetVLAD和OverlapNet，推理时间较DiSCO减少30.60%，模型参数仅为DiSCO的22.39%；在自采室内外数据集上集成到FAST-LIO2 SLAM系统，有效消除地图重影并修正z轴累积误差。未来计划引入3D注意力模型和视觉BEV信息进一步提升性能。实际应用中需注意方法依赖环境垂直结构和LiDAR垂直分辨率，对强俯仰/横滚误差或垂直视场差异较大的传感器可能敏感。\n\n### 关键要点\n\n- 提出Fast-DiSCO方法，将原始3D LiDAR点云预处理后投影为多层占用BEV图像，作为闭环检测的描述子输入。\n- 轻量级U形骨干网络在下采样中使用深度可分离卷积（DW-conv）和5×5大卷积核替代标准卷积与池化，以降低计算量和参数并扩大感受野。\n- 在上采样前引入基于深度可分离卷积的ASPP模块，从不同感受野提取多尺度BEV特征，增强点云特征信息。\n- 描述子通过傅里叶变换的幅值分量获得旋转不变性，核心公式为 S = |F_f ∘ F_c ∘ F_p|，并利用相位相关估计查询帧与历史帧之间的相对偏航角。\n- 相似度测量采用余弦距离计算1024维旋转不变描述子之间的距离，并通过Kd-Tree检索十个最相似候选，再结合几何验证确定闭环。\n- 在NCLT数据集上，使用2012-02-04训练，2012-03-17、2012-06-15、2012-08-20测试；在MulRan数据集上使用DDC03训练，DCC01和DCC02测试，BEV图像分辨率为40×120×20。\n- 在NCLT数据集上，Fast-DiSCO的Recall@1%为99.35%、Recall@1为91.23%，优于DiSCO的99.10%和89.47%；在MulRan数据集上Recall@1为63.02%，高于DiSCO的62.48%。\n- 在旋转后的NCLT数据集上，Fast-DiSCO的Recall@1%和Recall@1分别为96.12%和90.67%，验证了描述子的旋转不变性。\n- 偏航角估计方面，Fast-DiSCO在NCLT和MulRan数据集上的平均误差分别为1.26°和2.55°，标准差分别为1.85°和2.99%，优于DiSCO和OverlapNet。\n- 与DiSCO相比，Fast-DiSCO的推理时间减少30.60%，模型参数从7.77M压缩到1.74M（仅为22.39%），满足约10Hz LiDAR的实时性要求。\n- 将所提LCD方法集成到FAST-LIO2 SLAM系统后，在自采室内外数据集上消除了地图重影，并显著修正了z轴方向的累积定位误差。\n- 方法依赖环境的垂直结构和LiDAR垂直分辨率，对强俯仰/横滚误差、地形高程剧烈变化或垂直视场差异大的固态LiDAR可能性能下降。",
  "available_templates": [
    "deep_read",
    "simple"
  ]
}