作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2021, Vol. 47 ›› Issue (2): 279-284. doi: 10.19678/j.issn.1000-3428.0057147

• 图形图像处理 • 上一篇    下一篇

基于多模态特征融合的三维点云分类方法

顾砾, 季怡, 刘纯平   

  1. 苏州大学 计算机科学与技术学院, 江苏 苏州 215006
  • 收稿日期:2020-01-07 修回日期:2020-02-07 出版日期:2021-02-15 发布日期:2020-02-09
  • 作者简介:顾砾(1993-),女,硕士研究生,主研方向为图像处理、三维点云;季怡,副教授、博士;刘纯平,教授、博士、博士生导师。
  • 基金资助:
    国家自然科学基金(61972059,61773272);江苏省高等学校自然科学研究重大项目(19KJA230001)。

Classification Method of Three-Dimensional Point Cloud Based on Multiple Modal Feature Fusion

GU Li, JI Yi, LIU Chunping   

  1. School of Computer Science & Technology, Soochow University, Suzhou, Jiangsu 215006, China
  • Received:2020-01-07 Revised:2020-02-07 Online:2021-02-15 Published:2020-02-09

摘要: 针对点云数据本身信息量不足导致现有三维点云分类方法分类精度较低的问题,结合多模态特征融合,设计一种三维点云分类模型。通过引入投影图对点云数据信息进行扩充,将点云数据与图像数据同时作为输入,对PointCNN模型提取的点云特征与CNN模型提取的投影图特征进行加权融合,从而得到最终分类结果。在ModelNet40数据集上的分类结果表明,该模型的分类精度达到96.4%,相比PointCNN模型提升4.7个百分点。

关键词: 深度学习, 三维点云分类, PointCNN模型, 图像特征提取, 特征融合

Abstract: Existing three-dimensional point cloud classification methods suffer from low classification accuracy caused by the lack of information in point cloud data.To address the problem,this paper combines the multiple modal feature fusion to propose a three-dimensional point cloud classification model.This model expands the point cloud data information by introducing projection image,and takes point cloud data and image data as the input at the same time. The point cloud features extracted by PointCNN model and the projection image features extracted by CNN model are weighted and fused to obtain the final classification results.The classification results on the ModelNet40 dataset show that the classification accuracy of the proposed model reaches 96.4%,which is an increase of 4.7 percentage points compared with the PointCNN model.

Key words: deep learning, three-dimensional point cloud classification, PointCNN model, image feature extraction, feature fusion

中图分类号: