SCRnet: A Spatial Consistency Guided Network Using Contrastive Learning for Point Cloud Registration
Point cloud registration is used to find a rigid transformation from the source point cloud to the target point cloud. The main challenge in the point cloud registration is in finding correct correspondences in complex scenes that may contain many noise and repetitive structures. At present, many ex...
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2022-01-01
|
| Series: | Symmetry |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2073-8994/14/1/140 |
| _version_ | 1797490086049218560 |
|---|---|
| author | Huixiang Shao Zhijiang Zhang Xiaoyu Feng Dan Zeng |
| author_facet | Huixiang Shao Zhijiang Zhang Xiaoyu Feng Dan Zeng |
| author_sort | Huixiang Shao |
| collection | DOAJ |
| description | Point cloud registration is used to find a rigid transformation from the source point cloud to the target point cloud. The main challenge in the point cloud registration is in finding correct correspondences in complex scenes that may contain many noise and repetitive structures. At present, many existing methods use outlier rejections to help the network obtain more accurate correspondences, but they often ignore the spatial consistency between keypoints. Therefore, to address this issue, we propose a spatial consistency guided network using contrastive learning for point cloud registration (SCRnet), in which its overall stage is symmetrical. SCRnet consists of four blocks, namely feature extraction block, confidence estimation block, contrastive learning block and registration block. Firstly, we use <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mi>m</mi><mi>i</mi><mi>n</mi><mi>i</mi></mrow></semantics></math></inline-formula>-<inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mi>P</mi><mi>o</mi><mi>i</mi><mi>n</mi><mi>t</mi><mi>N</mi><mi>e</mi><mi>t</mi></mrow></semantics></math></inline-formula> to extract coarse local and global features. Secondly, we propose confidence estimation block, which formulate outlier rejection as confidence estimation problem of keypoint correspondences. In addition, the local spatial features are encoded into the confidence estimation block, which makes the correspondence possess local spatial consistency. Moreover, we propose contrastive learning block by constructing positive point pairs and hard negative point pairs and using Point-Pair-INfoNCE contrastive loss, which can further remove hard outliers through global spatial consistency. Finally, the proposed registration block selects a set of matching points with high spatial consistency and uses these matching sets to calculate multiple transformations, then the best transformation can be identified by initial alignment and Iterative Closest Point (ICP) algorithm. Extensive experiments are conducted on KITTI and nuScenes dataset, which demonstrate the high accuracy and strong robustness of SCRnet on point cloud registration task. |
| first_indexed | 2024-03-10T00:25:56Z |
| format | Article |
| id | doaj.art-aefcfe92d73d4314a05854ac3a844e05 |
| institution | Directory Open Access Journal |
| issn | 2073-8994 |
| language | English |
| last_indexed | 2024-03-10T00:25:56Z |
| publishDate | 2022-01-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Symmetry |
| spelling | doaj.art-aefcfe92d73d4314a05854ac3a844e052023-11-23T15:34:03ZengMDPI AGSymmetry2073-89942022-01-0114114010.3390/sym14010140SCRnet: A Spatial Consistency Guided Network Using Contrastive Learning for Point Cloud RegistrationHuixiang Shao0Zhijiang Zhang1Xiaoyu Feng2Dan Zeng3School of Communication and Information Engineering, Shanghai University, Shanghai 200444, ChinaSchool of Communication and Information Engineering, Shanghai University, Shanghai 200444, ChinaSchool of Communication and Information Engineering, Shanghai University, Shanghai 200444, ChinaSchool of Communication and Information Engineering, Shanghai University, Shanghai 200444, ChinaPoint cloud registration is used to find a rigid transformation from the source point cloud to the target point cloud. The main challenge in the point cloud registration is in finding correct correspondences in complex scenes that may contain many noise and repetitive structures. At present, many existing methods use outlier rejections to help the network obtain more accurate correspondences, but they often ignore the spatial consistency between keypoints. Therefore, to address this issue, we propose a spatial consistency guided network using contrastive learning for point cloud registration (SCRnet), in which its overall stage is symmetrical. SCRnet consists of four blocks, namely feature extraction block, confidence estimation block, contrastive learning block and registration block. Firstly, we use <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mi>m</mi><mi>i</mi><mi>n</mi><mi>i</mi></mrow></semantics></math></inline-formula>-<inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><mrow><mi>P</mi><mi>o</mi><mi>i</mi><mi>n</mi><mi>t</mi><mi>N</mi><mi>e</mi><mi>t</mi></mrow></semantics></math></inline-formula> to extract coarse local and global features. Secondly, we propose confidence estimation block, which formulate outlier rejection as confidence estimation problem of keypoint correspondences. In addition, the local spatial features are encoded into the confidence estimation block, which makes the correspondence possess local spatial consistency. Moreover, we propose contrastive learning block by constructing positive point pairs and hard negative point pairs and using Point-Pair-INfoNCE contrastive loss, which can further remove hard outliers through global spatial consistency. Finally, the proposed registration block selects a set of matching points with high spatial consistency and uses these matching sets to calculate multiple transformations, then the best transformation can be identified by initial alignment and Iterative Closest Point (ICP) algorithm. Extensive experiments are conducted on KITTI and nuScenes dataset, which demonstrate the high accuracy and strong robustness of SCRnet on point cloud registration task.https://www.mdpi.com/2073-8994/14/1/140point cloud registrationcontrastive learningspatial consistencydeep learning |
| spellingShingle | Huixiang Shao Zhijiang Zhang Xiaoyu Feng Dan Zeng SCRnet: A Spatial Consistency Guided Network Using Contrastive Learning for Point Cloud Registration Symmetry point cloud registration contrastive learning spatial consistency deep learning |
| title | SCRnet: A Spatial Consistency Guided Network Using Contrastive Learning for Point Cloud Registration |
| title_full | SCRnet: A Spatial Consistency Guided Network Using Contrastive Learning for Point Cloud Registration |
| title_fullStr | SCRnet: A Spatial Consistency Guided Network Using Contrastive Learning for Point Cloud Registration |
| title_full_unstemmed | SCRnet: A Spatial Consistency Guided Network Using Contrastive Learning for Point Cloud Registration |
| title_short | SCRnet: A Spatial Consistency Guided Network Using Contrastive Learning for Point Cloud Registration |
| title_sort | scrnet a spatial consistency guided network using contrastive learning for point cloud registration |
| topic | point cloud registration contrastive learning spatial consistency deep learning |
| url | https://www.mdpi.com/2073-8994/14/1/140 |
| work_keys_str_mv | AT huixiangshao scrnetaspatialconsistencyguidednetworkusingcontrastivelearningforpointcloudregistration AT zhijiangzhang scrnetaspatialconsistencyguidednetworkusingcontrastivelearningforpointcloudregistration AT xiaoyufeng scrnetaspatialconsistencyguidednetworkusingcontrastivelearningforpointcloudregistration AT danzeng scrnetaspatialconsistencyguidednetworkusingcontrastivelearningforpointcloudregistration |