V2PNet: Voxel-to-Point Feature Propagation and Fusion That Improves Feature Representation for Point Cloud Registration
Point-based and voxel-based methods can learn the local features of point clouds. However, although point-based methods are geometrically precise, the discrete nature of point clouds negatively affects feature learning performance. Moreover, although voxel-based methods can exploit the learning powe...
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2023-01-01
|
| Series: | IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/10130555/ |
| _version_ | 1797804856122015744 |
|---|---|
| author | Han Hu Yongkuo Hou Yulin Ding Guoqiang Pan Min Chen Xuming Ge |
| author_facet | Han Hu Yongkuo Hou Yulin Ding Guoqiang Pan Min Chen Xuming Ge |
| author_sort | Han Hu |
| collection | DOAJ |
| description | Point-based and voxel-based methods can learn the local features of point clouds. However, although point-based methods are geometrically precise, the discrete nature of point clouds negatively affects feature learning performance. Moreover, although voxel-based methods can exploit the learning power of convolutional neural networks, their resolution and detail extraction may be inadequate. Therefore, in this study, point-based and voxel-based methods are combined to enhance localization precision and matching distinctiveness. The core procedure is embodied in V2PNet, an innovative fused neural network that we design to perform voxel-to-pixel propagation and fusion, which seamlessly integrates the two encoder–decoder branches. Experiments are conducted on indoor and outdoor benchmark datasets with different platforms and sensors, i.e., the 3DMatch and Karlsruhe Institute of Technology and Toyota Technological Institute datasets, with the registration recall of 89.4% and the success rate of 99.86%, respectively. Qualitative and quantitative evaluations demonstrate that V2PNet has shown improvements in semantic awareness, geometric structure discernment, and other performance metrics. |
| first_indexed | 2024-03-13T05:43:17Z |
| format | Article |
| id | doaj.art-055f8b19670e451d8c795b4c9a62b82a |
| institution | Directory Open Access Journal |
| issn | 2151-1535 |
| language | English |
| last_indexed | 2024-03-13T05:43:17Z |
| publishDate | 2023-01-01 |
| publisher | IEEE |
| record_format | Article |
| series | IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing |
| spelling | doaj.art-055f8b19670e451d8c795b4c9a62b82a2023-06-13T23:00:21ZengIEEEIEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing2151-15352023-01-01165077508810.1109/JSTARS.2023.327883010130555V2PNet: Voxel-to-Point Feature Propagation and Fusion That Improves Feature Representation for Point Cloud RegistrationHan Hu0https://orcid.org/0000-0003-1137-2208Yongkuo Hou1https://orcid.org/0009-0009-4802-4320Yulin Ding2https://orcid.org/0000-0002-0539-1405Guoqiang Pan3Min Chen4https://orcid.org/0000-0003-1381-7290Xuming Ge5https://orcid.org/0000-0002-1032-1938Faculty of Geosciences and Environmental Engineering, Southwest Jiaotong University, Chengdu, Sichuan, ChinaFaculty of Geosciences and Environmental Engineering, Southwest Jiaotong University, Chengdu, Sichuan, ChinaFaculty of Geosciences and Environmental Engineering, Southwest Jiaotong University, Chengdu, Sichuan, ChinaEquipment Project Management Center, Chinese People's Armed Police Force, Beijing, ChinaFaculty of Geosciences and Environmental Engineering, Southwest Jiaotong University, Chengdu, Sichuan, ChinaFaculty of Geosciences and Environmental Engineering, Southwest Jiaotong University, Chengdu, Sichuan, ChinaPoint-based and voxel-based methods can learn the local features of point clouds. However, although point-based methods are geometrically precise, the discrete nature of point clouds negatively affects feature learning performance. Moreover, although voxel-based methods can exploit the learning power of convolutional neural networks, their resolution and detail extraction may be inadequate. Therefore, in this study, point-based and voxel-based methods are combined to enhance localization precision and matching distinctiveness. The core procedure is embodied in V2PNet, an innovative fused neural network that we design to perform voxel-to-pixel propagation and fusion, which seamlessly integrates the two encoder–decoder branches. Experiments are conducted on indoor and outdoor benchmark datasets with different platforms and sensors, i.e., the 3DMatch and Karlsruhe Institute of Technology and Toyota Technological Institute datasets, with the registration recall of 89.4% and the success rate of 99.86%, respectively. Qualitative and quantitative evaluations demonstrate that V2PNet has shown improvements in semantic awareness, geometric structure discernment, and other performance metrics.https://ieeexplore.ieee.org/document/10130555/3-D feature point3-D metric learningpoint cloud registrationsparse convolution |
| spellingShingle | Han Hu Yongkuo Hou Yulin Ding Guoqiang Pan Min Chen Xuming Ge V2PNet: Voxel-to-Point Feature Propagation and Fusion That Improves Feature Representation for Point Cloud Registration IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing 3-D feature point 3-D metric learning point cloud registration sparse convolution |
| title | V2PNet: Voxel-to-Point Feature Propagation and Fusion That Improves Feature Representation for Point Cloud Registration |
| title_full | V2PNet: Voxel-to-Point Feature Propagation and Fusion That Improves Feature Representation for Point Cloud Registration |
| title_fullStr | V2PNet: Voxel-to-Point Feature Propagation and Fusion That Improves Feature Representation for Point Cloud Registration |
| title_full_unstemmed | V2PNet: Voxel-to-Point Feature Propagation and Fusion That Improves Feature Representation for Point Cloud Registration |
| title_short | V2PNet: Voxel-to-Point Feature Propagation and Fusion That Improves Feature Representation for Point Cloud Registration |
| title_sort | v2pnet voxel to point feature propagation and fusion that improves feature representation for point cloud registration |
| topic | 3-D feature point 3-D metric learning point cloud registration sparse convolution |
| url | https://ieeexplore.ieee.org/document/10130555/ |
| work_keys_str_mv | AT hanhu v2pnetvoxeltopointfeaturepropagationandfusionthatimprovesfeaturerepresentationforpointcloudregistration AT yongkuohou v2pnetvoxeltopointfeaturepropagationandfusionthatimprovesfeaturerepresentationforpointcloudregistration AT yulinding v2pnetvoxeltopointfeaturepropagationandfusionthatimprovesfeaturerepresentationforpointcloudregistration AT guoqiangpan v2pnetvoxeltopointfeaturepropagationandfusionthatimprovesfeaturerepresentationforpointcloudregistration AT minchen v2pnetvoxeltopointfeaturepropagationandfusionthatimprovesfeaturerepresentationforpointcloudregistration AT xumingge v2pnetvoxeltopointfeaturepropagationandfusionthatimprovesfeaturerepresentationforpointcloudregistration |