DR_PVNet: An Improved PVNet Network for Weakly Textured Object Pose Estimation in Indoor Complex Environments
6D pose estimation of objects is an essential task in computer vision. However, estimating the 6D pose of a weakly textured object from an image is challenging due to the occluded environment and the weakly textured appearance. PVNet locates the 2D keypoints by obtaining the segmentation results of...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/10855419/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1832575579247869952 |
|---|---|
| author | Mengru Niu Changyong Li |
| author_facet | Mengru Niu Changyong Li |
| author_sort | Mengru Niu |
| collection | DOAJ |
| description | 6D pose estimation of objects is an essential task in computer vision. However, estimating the 6D pose of a weakly textured object from an image is challenging due to the occluded environment and the weakly textured appearance. PVNet locates the 2D keypoints by obtaining the segmentation results of the object and the prediction results of the dense unit vector field of the keypoints, followed by a voting strategy. This method shows some robustness to weakly textured object pose estimation. In order to enhance the precision of target object segmentation and the efficacy of the voting scheme, this paper builds upon the PVNet model to develop DR_PVNet, which is capable of performing pose estimation for weakly textured objects in complex environments. The first step is to add the RFB_a module between the encoder and decoder, enhancing the semantic segmentation of the network. Secondly, a distance filtering scheme is incorporated into the keypoint voting scheme to force pixels that are further away from the keypoint hypotheses to improve the vote quality. The DR_PVNet algorithm achieves a 2D projection metric of 99.38% and an ADD(-S) metric of 90.12 % on the weakly textured public dataset LINEMOD. Finally, the methodology of this paper is extended to the pose estimation of weakly textured objects in everyday life using a dataset production method rendered by BlenderProc. The 2D projection metric reached 94.95%, and the ADD(-S) metric reached 91.65 % for the self-made dataset. |
| format | Article |
| id | doaj-art-422e3564f49247ba9c2cbb6198cedab7 |
| institution | Kabale University |
| issn | 2169-3536 |
| language | English |
| publishDate | 2025-01-01 |
| publisher | IEEE |
| record_format | Article |
| series | IEEE Access |
| spelling | doaj-art-422e3564f49247ba9c2cbb6198cedab72025-01-31T23:04:31ZengIEEEIEEE Access2169-35362025-01-0113195761958710.1109/ACCESS.2025.353485610855419DR_PVNet: An Improved PVNet Network for Weakly Textured Object Pose Estimation in Indoor Complex EnvironmentsMengru Niu0https://orcid.org/0009-0008-0714-2913Changyong Li1College of Intelligent Manufacturing Modern Industry, Xinjiang University, Urumqi, Xinjiang, ChinaCollege of Intelligent Manufacturing Modern Industry, Xinjiang University, Urumqi, Xinjiang, China6D pose estimation of objects is an essential task in computer vision. However, estimating the 6D pose of a weakly textured object from an image is challenging due to the occluded environment and the weakly textured appearance. PVNet locates the 2D keypoints by obtaining the segmentation results of the object and the prediction results of the dense unit vector field of the keypoints, followed by a voting strategy. This method shows some robustness to weakly textured object pose estimation. In order to enhance the precision of target object segmentation and the efficacy of the voting scheme, this paper builds upon the PVNet model to develop DR_PVNet, which is capable of performing pose estimation for weakly textured objects in complex environments. The first step is to add the RFB_a module between the encoder and decoder, enhancing the semantic segmentation of the network. Secondly, a distance filtering scheme is incorporated into the keypoint voting scheme to force pixels that are further away from the keypoint hypotheses to improve the vote quality. The DR_PVNet algorithm achieves a 2D projection metric of 99.38% and an ADD(-S) metric of 90.12 % on the weakly textured public dataset LINEMOD. Finally, the methodology of this paper is extended to the pose estimation of weakly textured objects in everyday life using a dataset production method rendered by BlenderProc. The 2D projection metric reached 94.95%, and the ADD(-S) metric reached 91.65 % for the self-made dataset.https://ieeexplore.ieee.org/document/10855419/6D pose estimationweak texturesemantic segmentationvoting |
| spellingShingle | Mengru Niu Changyong Li DR_PVNet: An Improved PVNet Network for Weakly Textured Object Pose Estimation in Indoor Complex Environments IEEE Access 6D pose estimation weak texture semantic segmentation voting |
| title | DR_PVNet: An Improved PVNet Network for Weakly Textured Object Pose Estimation in Indoor Complex Environments |
| title_full | DR_PVNet: An Improved PVNet Network for Weakly Textured Object Pose Estimation in Indoor Complex Environments |
| title_fullStr | DR_PVNet: An Improved PVNet Network for Weakly Textured Object Pose Estimation in Indoor Complex Environments |
| title_full_unstemmed | DR_PVNet: An Improved PVNet Network for Weakly Textured Object Pose Estimation in Indoor Complex Environments |
| title_short | DR_PVNet: An Improved PVNet Network for Weakly Textured Object Pose Estimation in Indoor Complex Environments |
| title_sort | dr pvnet an improved pvnet network for weakly textured object pose estimation in indoor complex environments |
| topic | 6D pose estimation weak texture semantic segmentation voting |
| url | https://ieeexplore.ieee.org/document/10855419/ |
| work_keys_str_mv | AT mengruniu drpvnetanimprovedpvnetnetworkforweaklytexturedobjectposeestimationinindoorcomplexenvironments AT changyongli drpvnetanimprovedpvnetnetworkforweaklytexturedobjectposeestimationinindoorcomplexenvironments |