PPE: Point position embedding for single object tracking in point clouds
Abstract Existing 3D single object tracking methods primarily extract features from the global coordinates of point clouds, overlooking the potential exploitation of their positional information. However, due to the unordered, sparse, and irregular nature of point clouds, effectively exploring their...
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Wiley
2023-08-01
|
| Series: | Electronics Letters |
| Subjects: | |
| Online Access: | https://doi.org/10.1049/ell2.12914 |
| _version_ | 1797745134409875456 |
|---|---|
| author | Yuanzhi Su Yuan‐Gen Wang Weijia Wang Guopu Zhu |
| author_facet | Yuanzhi Su Yuan‐Gen Wang Weijia Wang Guopu Zhu |
| author_sort | Yuanzhi Su |
| collection | DOAJ |
| description | Abstract Existing 3D single object tracking methods primarily extract features from the global coordinates of point clouds, overlooking the potential exploitation of their positional information. However, due to the unordered, sparse, and irregular nature of point clouds, effectively exploring their positional information presents a significant challenge. In this letter, the network is explicitly reformulated by introducing a point position embedding module in conjunction with a self‐attention coding module, replacing the use of global coordinate inputs. The proposed reformulation is further integrated into a top‐notch model M2‐Track, called Point Position Embedding (PPE) in this letter. Comprehensive empirical analysis are performed on the KITTI and NuScenes datasets. Experimental results show that the PPE surpasses M2‐Track by a large margin in overall performance. Especially for the challenging NuScenes dataset, the method attains the highest precision and success in all classes compared to state‐of‐the‐art methods. The code is available at https://github.com/GZHU‐DVL/PPE. |
| first_indexed | 2024-03-12T15:19:08Z |
| format | Article |
| id | doaj.art-a1e86da11bba47928219beb94e6f26c1 |
| institution | Directory Open Access Journal |
| issn | 0013-5194 1350-911X |
| language | English |
| last_indexed | 2024-03-12T15:19:08Z |
| publishDate | 2023-08-01 |
| publisher | Wiley |
| record_format | Article |
| series | Electronics Letters |
| spelling | doaj.art-a1e86da11bba47928219beb94e6f26c12023-08-11T07:18:29ZengWileyElectronics Letters0013-51941350-911X2023-08-015915n/an/a10.1049/ell2.12914PPE: Point position embedding for single object tracking in point cloudsYuanzhi Su0Yuan‐Gen Wang1Weijia Wang2Guopu Zhu3School of Computer Science and Cyber Engineering Guangzhou University GuangzhouChinaSchool of Computer Science and Cyber Engineering Guangzhou University GuangzhouChinaSchool of Information Technology Deakin University Waurn Ponds CampusGeelongAustraliaSchool of Cyberspace SecurityHarbin Institute of TechnologyHarbinChinaAbstract Existing 3D single object tracking methods primarily extract features from the global coordinates of point clouds, overlooking the potential exploitation of their positional information. However, due to the unordered, sparse, and irregular nature of point clouds, effectively exploring their positional information presents a significant challenge. In this letter, the network is explicitly reformulated by introducing a point position embedding module in conjunction with a self‐attention coding module, replacing the use of global coordinate inputs. The proposed reformulation is further integrated into a top‐notch model M2‐Track, called Point Position Embedding (PPE) in this letter. Comprehensive empirical analysis are performed on the KITTI and NuScenes datasets. Experimental results show that the PPE surpasses M2‐Track by a large margin in overall performance. Especially for the challenging NuScenes dataset, the method attains the highest precision and success in all classes compared to state‐of‐the‐art methods. The code is available at https://github.com/GZHU‐DVL/PPE.https://doi.org/10.1049/ell2.12914computer visionimage motion analysisimage recognitionobject tracking |
| spellingShingle | Yuanzhi Su Yuan‐Gen Wang Weijia Wang Guopu Zhu PPE: Point position embedding for single object tracking in point clouds Electronics Letters computer vision image motion analysis image recognition object tracking |
| title | PPE: Point position embedding for single object tracking in point clouds |
| title_full | PPE: Point position embedding for single object tracking in point clouds |
| title_fullStr | PPE: Point position embedding for single object tracking in point clouds |
| title_full_unstemmed | PPE: Point position embedding for single object tracking in point clouds |
| title_short | PPE: Point position embedding for single object tracking in point clouds |
| title_sort | ppe point position embedding for single object tracking in point clouds |
| topic | computer vision image motion analysis image recognition object tracking |
| url | https://doi.org/10.1049/ell2.12914 |
| work_keys_str_mv | AT yuanzhisu ppepointpositionembeddingforsingleobjecttrackinginpointclouds AT yuangenwang ppepointpositionembeddingforsingleobjecttrackinginpointclouds AT weijiawang ppepointpositionembeddingforsingleobjecttrackinginpointclouds AT guopuzhu ppepointpositionembeddingforsingleobjecttrackinginpointclouds |