计算机工程与应用 ›› 2020, Vol. 56 ›› Issue (1): 38-45.DOI: 10.3778/j.issn.1002-8331.1910-0157

• 热点与综述 • 上一篇    下一篇

基于深度学习的点云分割方法综述

俞斌,董晨,刘延华,程烨   

  1. 1.福州大学 数学与计算机科学学院,福州 350116
    2.福州大学 福建省网络计算与智能信息处理重点实验室,福州 350116
    3.福州大学 网络系统信息安全福建省高校重点实验室,福州 350116
  • 出版日期:2020-01-01 发布日期:2020-01-02

Deep Learning Based Point Cloud Segmentation: A Survey

YU Bin, DONG Chen, LIU Yanhua, CHENG Ye   

  1. 1.College of Mathematics and Computer Science, Fuzhou University, Fuzhou 350116, China
    2.Fujian Provincial Key Laboratory of Network Computing and Intelligent Information Processing, Fuzhou University, Fuzhou 350116, China
    3.Key Lab of Information Security of Network Systems, Fuzhou University, Fuzhou 350116, China
  • Online:2020-01-01 Published:2020-01-02

摘要: 点云分割是点云数据理解中的一个关键技术,但传统算法无法进行实时语义分割。近年来深度学习被应用在点云分割上并取得了重要进展。综述了近四年来基于深度学习的点云分割的最新工作,按基本思想分为基于视图和投影的方法、基于体素的方法、无序点云的方法、有序点云的方法以及无监督学习的方法,并简要评述;最后分析各类方法优劣并展望未来研究趋势。

关键词: 深度学习, 点云标注, 语义分割

Abstract: Point cloud segmentation is a key technology in point cloud data understanding, but traditional algorithms cannot perform real-time semantic segmentation. In recent years, deep learning has applied on point cloud segmentation and achieved significant progress. This paper reviews the latest work of point cloud segmentation based on deep learning in the past four years. In line with main content, it is divided into five kind of methods: view-based and projection-based method, volumetric method, unordered point cloud method, ordered point cloud method and unsupervised deep learning method, meanwhile it givs a brief review. Finally, the paper analyzes the advantages and disadvantages of various methods and looks forward to future research trends.

Key words: deep learning, point cloud labeling, semantic segmentation