Spherical coordinate transformation-embedded deep network for primitive instance segmentation of point clouds

被引:5
作者
Li, Wei [1 ,2 ]
Xie, Sijing [1 ,2 ]
Min, Weidong [1 ,2 ]
Jiang, Yifei [1 ,2 ]
Wang, Cheng [3 ]
Li, Jonathan [4 ]
机构
[1] Nanchang Univ, Sch Software, 235 Nanjing Rd East, Nanchang 330047, Jiangxi, Peoples R China
[2] Nanchang Univ, Jiangxi Key Lab Smart City, Nanchang, Peoples R China
[3] Xiamen Univ, Sch Informat Sci & Engn, Fujian Key Lab Sensing & Comp Smart Cities, 422 Siming Rd South, Xiamen 361005, Fujian, Peoples R China
[4] Univ Waterloo, Dept Geog & Environm Management & Syst Design Engn, 200 Univ Ave West, Waterloo, ON N21 3G1, Canada
基金
中国国家自然科学基金;
关键词
Primitive instance segmentation; Spherical coordinate transformation; Relation matrix; RECONSTRUCTION;
D O I
10.1016/j.jag.2022.102983
中图分类号
TP7 [遥感技术];
学科分类号
081102 ; 0816 ; 081602 ; 083002 ; 1404 ;
摘要
In this research, a primitive prediction network embedding Spherical Coordinate Transformation (named SCT-Net), which is a simple and end-to-end deep neural network, is proposed for primitive instance segmentation of point clouds. The key point of SCT-Net is to excavate the relationship between local neighborhood points. First, in order to enhance the compacted expression of local feature, a spherical coordinate transformation is embedded to a deep network. Second, the embedded network is constructed to predict the point grouping proposals and classify the primitives corresponding to each proposal, which can segment primitive instance directly. Third, the feature relationship between each two points is revealed by the constructed relation matrix. The designed loss function not only encourages the embedded network to describe local surface properties, but also produces a grouping strategy accurately for each point. Experiments show that the proposed SCT-Net achieves the state-of-the-art performance than representative methods. At the same time, the capability of spherical coordinate transformation has been demonstrated to improve primitive instance segmentation.
引用
收藏
页数:8
相关论文
共 37 条
[31]   Learning Shape Abstractions by Assembling Volumetric Primitives [J].
Tulsiani, Shubham ;
Su, Hao ;
Guibas, Leonidas J. ;
Efros, Alexei A. ;
Malik, Jitendra .
30TH IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2017), 2017, :1466-1474
[32]   SGPN: Similarity Group Proposal Network for 3D Point Cloud Instance Segmentation [J].
Wang, Weiyue ;
Yu, Ronald ;
Huang, Qiangui ;
Neumann, Ulrich .
2018 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR), 2018, :2569-2578
[33]   Constructing 3D CSG Models from 3D Raw Point Clouds [J].
Wu, Q. ;
Xu, K. ;
Wang, J. .
COMPUTER GRAPHICS FORUM, 2018, 37 (05) :221-232
[34]   SUN3D: A Database of Big Spaces Reconstructed using SfM and Object Labels [J].
Xiao, Jianxiong ;
Owens, Andrew ;
Torralba, Antonio .
2013 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV), 2013, :1625-1632
[35]   An Optimal Hierarchical Clustering Approach to Mobile LiDAR Point Clouds [J].
Xu, Sheng ;
Wang, Ruisheng ;
Wang, Hao ;
Zheng, Han .
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, 2020, 21 (07) :2765-2776
[36]   HPNet: Deep Primitive Segmentation Using Hybrid Representations [J].
Yan, Siming ;
Yang, Zhenpei ;
Ma, Chongyang ;
Huang, Haibin ;
Vouga, Etienne ;
Huang, Qixing .
2021 IEEE/CVF INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV 2021), 2021, :2733-2742
[37]   3D-PRNN: Generating Shape Primitives with Recurrent Neural Networks [J].
Zou, Chuhang ;
Yumer, Ersin ;
Yang, Jimei ;
Ceylan, Duygu ;
Hoiem, Derek .
2017 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV), 2017, :900-909