深度学习驱动的点云上采样研究综述  

Survey of Point Cloud Upsampling Based on Deep Learning

作　　者：韩亚振 尹梦晓[1,2] 杨锋 钟诚[1,2,3] HAN Yazhen;YIN Mengxiao;YANG Feng;ZHONG Cheng(School of Computer and Electronics Information,Guangxi University,Nanning 530004,China;Guangxi Key Laboratory of Multimedia Communications and Network Technology,Nanning 530004,China;Key Laboratory of Parallel,Distributed and Intelligent Computing(Guangxi University),Education Department of Guangxi Zhuang AutonomousRegion,Nanning 530004,China)

机构地区：[1]广西大学计算机与电子信息学院,南宁530004 [2]广西多媒体通信与网络技术重点实验室,南宁530004 [3]广西壮族自治区教育厅并行分布与智能计算重点实验室(广西大学),南宁530004

出　　处：《小型微型计算机系统》2025年第3期645-654,共10页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61762007)资助.

摘　　要：由三维扫描设备得到的点云经常是稀疏的、不均匀的,甚至因为遮挡而丢失数据,因此,点云上采样在渲染、三维重建等领域扮演了越来越重要的角色.随着深度学习的不断发展,基于神经网络的点云上采样方法逐渐成为该方向的主流技术.本文系统的阐述了目前主要的基于深度学习的点云上采样方法,首先介绍了三段式点云上采样、基于离散微分几何的点云上采样以及基于生成模型的点云上采样.随后讨论了无监督式点云上采样和任意倍率点云上采样.最后介绍了相关数据集和评估指标.文章分析了各类方法的特点、优势和挑战,并讨论了未来研究的潜在方向和应用前景,希望能为研究人员提供较为全面的参考.The point cloud obtained by 3D scanning equipment is often sparse,uneven,and even loses data due to occlusion.Therefore,point cloud upsampling plays an increasingly important role in fields such as rendering and 3D reconstruction.With the continuous development of deep learning,point cloud upsampling methods based on neural networks have gradually become the mainstream technology in this area.This paper systematically expounds on the current major deep learning-based point cloud upsampling methods.First,it introduces three-segment point cloud upsampling,point cloud upsampling based on discrete differential geometry,and point cloud upsampling based on generative models.It then discusses unsupervised point cloud upsampling and arbitrary-ratio point cloud upsampling.Finally,it presents relevant datasets and evaluation metrics.This article analyzes the characteristics,advantages and challenges of various methods,and discusses the potential directions and application prospects of future research,hoping to provide a more comprehensive reference for researchers.

关 键 词：点云 点云上采样 深度学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]