[1]黄 丽,刘心维,肖 建*.基于深度学习的高精度点云补全算法[J].计算机技术与发展,2023,33(04):62-68.[doi:10. 3969 / j. issn. 1673-629X. 2023. 04. 009]
　HUANG Li,LIU Xin-wei,XIAO Jian*.High Precision Point Cloud Completion Algorithm Based on Deep Learning[J].,2023,33(04):62-68.[doi:10. 3969 / j. issn. 1673-629X. 2023. 04. 009]
点击复制

基于深度学习的高精度点云补全算法()
分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
33
期数:
2023年04期
页码:
62-68
栏目:
媒体计算
出版日期:
2023-04-10

文章信息/Info

Title:
High Precision Point Cloud Completion Algorithm Based on Deep Learning
文章编号:
1673-629X(2023)04-0062-07
作者:
黄 丽; 刘心维; 肖 建*
南京邮电大学 电子与光学工程学院、柔性电子学院,江苏 南京 210046
Author(s):
HUANG Li; LIU Xin-wei; XIAO Jian*
School of Electronic and Optical Engineering and School of Flexible Electronics,Nanjing University of Posts and Telecommunications,Nanjing 210046,China
关键词:
点云补全; 三维点云; 特征提取; 卷积; 深度学习
Keywords:
point cloud completion; 3D point cloud; feature extraction; convolution; deep learning
分类号:
TP391
DOI:
10. 3969 / j. issn. 1673-629X. 2023. 04. 009
摘要:
由于获取三维点云数据设备分辨率低及物体遮挡等原因,不可避免地会导致获取的点云数据通常是不完整的,这对点云分割、点云重建等下游任务会产生巨大的影响。 近年来基于深度学习的点云补全算法开始广泛应用并取得较好的效果,但大多数算法只关注点云的全局信息,并没有充分考虑到点云的局部特征,难以表征点云空间复杂的变化关系,所以补全精度上还存在欠缺。 该文提出一种基于深度学习的高精度点云补全算法,算法整体采用编码器-解码器结构,创造性地在特征提取模块引入卷积层 DOConv,并在特征融合模块添加结合空间注意力机制和通道注意力机制的双重注意力机制,可以融合不同层次的特征。 该算法可以灵活地对点云全局和局部特征进行提取并综合关键点的局部关联性和全局信息。 实验结果表明,与几个主流算法相比,该算法补全精度更高,可以得到更为完整准确的点云模型。
Abstract:
Due to the low resolution of the device for acquiring 3D point cloud data and the occlusion of objects,the acquired point clouddata is usually incomplete,which will have a huge impact on downstream tasks such as? point cloud segmentation and point cloud reconstruction. In recent years,point cloud completion algorithms based on deep learning have been widely used and achieved good results,butmost algorithms only focus on the global information of point clouds,and do not fully consider the local features of point clouds,makingit difficult to characterize the point cloud space. There is a complex change relationship,so there is still a lack of completion accuracy.We propose a high - precision point cloud completion algorithm based on deep learning. The algorithm adopts the encoder - decoderstructure as a whole,creatively introduces the convolutional layer DOConv in the feature extraction module,and adds combined spatial attention in the feature fusion module. The dual attention mechanism of force mechanism and channel attention mechanism can fusefeatures at different levels. The algorithm can flexibly extract global and local features of point clouds and synthesize local correlation andglobal information of key points. The experimental results show that compared with several mainstream algorithms, the proposedalgorithm has higher completion accuracy and can obtain a more complete and accurate point cloud model.

相似文献/References:

[1]周勇飞,徐昱琳,吕晓梦,等.基于双目的三维点云数据的获取与预处理[J].计算机技术与发展,2014,24(03):22.
　ZHOU Yong-fei,XU Yu-lin,Lü Xiao-meng,et al.Three-dimensional Point Cloud Data Acquisition and Pre-processing Based on Binocular[J].,2014,24(04):22.
[2]李 昭,宋 壹,陈 鹏.基于监督学习的数据预测服务构建方法[J].计算机技术与发展,2019,29(09):188.[doi:10. 3969 / j. issn. 1673-629X. 2019. 09. 036]
　LI Zhao,SONG Yi,CHEN Peng.Data Prediction Service Construction Based on Supervised Learning[J].,2019,29(04):188.[doi:10. 3969 / j. issn. 1673-629X. 2019. 09. 036]
[3]荆树旭,赵 娇*.基于弱随机相机位姿图像的三维场景恢复[J].计算机技术与发展,2023,33(05):194.[doi:10. 3969 / j. issn. 1673-629X. 2023. 05. 029]
　JING Shu-xu,ZHAO Jiao*.3D Scene Restoration Based on Weak Random Camera Pose Image[J].,2023,33(04):194.[doi:10. 3969 / j. issn. 1673-629X. 2023. 05. 029]

更新日期/Last Update: 2023-04-10