《计算机应用》唯一官方网站 ›› 2025, Vol. 45 ›› Issue (10): 3294-3301.DOI: 10.11772/j.issn.1001-9081.2024101437

• 多媒体计算与计算机仿真 • 上一篇    

基于自适应邻域特征融合的多阶段点云补全网络

李维刚1,2, 曹文杰1(), 李金灵1   

  1. 1.武汉科技大学 信息科学与工程学院,武汉 430081
    2.冶金自动化与检测技术教育部工程研究中心(武汉科技大学),武汉 430081
  • 收稿日期:2024-10-12 修回日期:2024-11-28 接受日期:2024-12-02 发布日期:2024-12-05 出版日期:2025-10-10
  • 通讯作者: 曹文杰
  • 作者简介:李维刚(1977—),男,湖北咸宁人,教授,博士,主要研究方向:工业过程控制、人工智能、机器学习
    曹文杰(2000—),男,湖北孝感人,硕士研究生,主要研究方向:点云重建、点云分割 Email:wenjccc@163.com
    李金灵(1996—),男,湖北咸宁人,博士研究生,主要研究方向:计算机视觉。
  • 基金资助:
    湖北省科技人才服务企业项目(202400288)

Multi-stage point cloud completion network based on adaptive neighborhood feature fusion

Weigang LI1,2, Wenjie CAO1(), Jinling LI1   

  1. 1.School of Information Science and Engineering,Wuhan University of Science and Technology,Wuhan Hubei 430081,China
    2.Engineering Research Center for Metallurgical Automation and Measurement Technology of Ministry of Education (Wuhan University of Science and Technology),Wuhan Hubei 430081,China
  • Received:2024-10-12 Revised:2024-11-28 Accepted:2024-12-02 Online:2024-12-05 Published:2025-10-10
  • Contact: Wenjie CAO
  • About author:LI Weigang, born in 1977, Ph. D., professor. His research interests include industrial process control, artificial intelligence, machine learning.
    CAO Wenjie, born in 2000, M. S. candidate. His research interests include point cloud reconstruction, point cloud segmentation.
    LI Jinling, born in 1996, Ph. D. candidate. His research interests include computer vision.
  • Supported by:
    Hubei Provincial Science and Technology Talent Serving Enterprise Project(202400288)

摘要:

点云补全指利用不完整的点云数据重建高质量的完整点云。然而,现有的大多数点云补全网络在捕捉局部特征和重建细节方面存在不足,导致生成的点云在局部细节和补全精度上表现不佳。为解决上述问题,提出一种基于自适应邻域特征融合的多阶段点云补全网络(ANFF-Net)。首先,特征提取器通过自适应调整关键点的邻域选择适应不同形状的点云,有效捕捉不同语义点之间的空间关系,减少局部细节信息的丢失;其次,特征拓展器利用局部感知Transformer进一步扩展邻近点的局部特征信息,提升网络的细节恢复能力;最后,点云生成器采用交叉注意力机制选择性传递不完整点云的局部特征信息,并使用折叠模块逐步细化点云的局部区域,显著增强补全后点云的细节保留,生成更一致的几何细节。实验结果表明,ANFF-Net在ShapeNet55数据集上的平均补全精度相较于ProxyFormer提升了9.68%,并在PCN和KITTI数据集上取得了较好的补全效果。可视化结果显示,ANFF-Net生成的点云具有更高的细粒度,形状更接近真实值。

关键词: 点云补全, 局部特征, 自适应邻域, 局部感知, 交叉注意力

Abstract:

Point cloud completion aims to reconstruct a high-quality complete point cloud from incomplete point cloud data. However, most existing point cloud completion networks have limitations in capturing local features and reconstructing details, resulting in poor performance of the generated point cloud in terms of local details and completion accuracy. To address these issues, a multi-stage point cloud completion Network based on Adaptive Neighborhood Feature Fusion (ANFF-Net) was proposed. Firstly, the neighborhood selection of key points was adjusted by the feature extractor adaptively to adapt to different shapes of point clouds, so as to capture spatial relationships between points with different semantics effectively, thereby reducing loss of the local details. Then, a local perception Transformer was used by the feature expander to further expand local feature information of the neighboring points, thereby improving the network’s ability to recover details. Finally, a cross-attention mechanism was applied by the point cloud generator to propagate local feature information of the incomplete point cloud selectively, and a folding module was used to refine the local regions gradually, thereby enhancing detail retention of the completed point cloud significantly and generating more consistent geometric details. Experimental results show that ANFF-Net improves the average completion accuracy by 9.68% compared to ProxyFormer on the ShapeNet55 dataset and achieves good completion performance on the PCN and KITTI datasets. Visualization results indicate that the point clouds generated by ANFF-Net have finer granularity and are closer to the ground truth in shape.

Key words: point cloud completion, local feature, adaptive neighborhood, local perception, cross-attention

中图分类号: