中国科学院机构知识库网格
Chinese Academy of Sciences Institutional Repositories Grid
首页 机构 成果 学者
  
  1. CAS IR Grid
  2. 地理科学与资源研究所
  3. 中国科学院地理科学与资源研究所
  4. 资源与环境信息系统国家重点实验室
  5. 外文论文
LiGAPU: A LiDAR point cloud upsampling network for multiple complex scenes

文献类型：期刊论文

作者Fu, Bolin2; Sui, Mingzhe2; Li, Huajian2; Yang, Fei1; Yao, Hang2; Deng, Tengfang2; Zhang, Xing2
刊名PATTERN RECOGNITION
出版日期2026-02-01
卷号170页码:112132
关键词LiDAR point cloud Arbitrary upsampling rates Dual-stream convolution Point-to-point distance prediction Multiple scenes validation Deep learning
ISSN号0031-3203
DOI10.1016/j.patcog.2025.112132
产权排序2
文献子类Article
英文摘要Point cloud upsampling is essential for restoring fine details in point clouds. While most existing methods focus on single-object point clouds with regular geometric shapes, there is limited research on upsampling LiDAR point clouds from complex scenes, which often exhibit intricate structures, non-uniform density, and significant noise. To fill this gap, we propose LiGAPU, a novel Geometry-Aware LiDAR point cloud upsampling network designed to enhance the modeling of spatial structures in these complex scenes. The core of LiGAPU lies in two key modules: a Dual-stream Gating mechanism-based convolutional (DGConv) module that robustly captures both scalar and vector features while modeling their global dependencies, and a GCS module for precise point-to-point distance prediction. By combining 3D Gaussian feature interpolation with Channel-Spatial attention mechanisms, the GCS module effectively models fine-grained local relationships to improve the positional accuracy of upsampled points. We conduct experiments on public benchmarks as well as our self-collected datasets. Experimental results demonstrate that LiGAPU improves upsampling accuracy across all datasets while preserving structural consistency, and exhibits strong robustness and generalization in complex scenes. The code is available at https://github.com/Sailkiki/LiGAPU.
URL标识查看原文
WOS研究方向Computer Science ; Engineering
语种英语
WOS记录号WOS:001573119000001
出版者ELSEVIER SCI LTD
源URL[http://ir.igsnrr.ac.cn/handle/311030/216021]  
专题资源与环境信息系统国家重点实验室_外文论文
通讯作者Fu, Bolin
作者单位1.Chinese Acad Sci, Inst Geog Sci & Nat Resources Res, State Key Lab Resources & Environm Informat Syst, Beijing 100101, Peoples R China
2.Guilin Univ Technol, Coll Geomat & Geoinformat, Guilin 541006, Peoples R China;
推荐引用方式
GB/T 7714		 Fu, Bolin,Sui, Mingzhe,Li, Huajian,et al. LiGAPU: A LiDAR point cloud upsampling network for multiple complex scenes[J]. PATTERN RECOGNITION,2026,170:112132.
APA Fu, Bolin.,Sui, Mingzhe.,Li, Huajian.,Yang, Fei.,Yao, Hang.,...&Zhang, Xing.(2026).LiGAPU: A LiDAR point cloud upsampling network for multiple complex scenes.PATTERN RECOGNITION,170,112132.
MLA Fu, Bolin,et al."LiGAPU: A LiDAR point cloud upsampling network for multiple complex scenes".PATTERN RECOGNITION 170(2026):112132.

入库方式： OAI收割

来源：地理科学与资源研究所

浏览0
下载0
收藏0
其他版本

除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。