Salient sparse visual odometry with pose-only supervision
Visual Odometry (VO) is vital for the navigation of autonomous systems, providing accurate position and orientation estimates at reasonable costs. While traditional VO methods excel in some conditions, they struggle with challenges like variable lighting and motion blur. Deep learning-based VO,...
| Main Authors: | , , , , , |
|---|---|
| Other Authors: | |
| Format: | Journal Article |
| Language: | English |
| Published: |
2024
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/178995 http://arxiv.org/abs/2404.04677v1 |
| _version_ | 1826115365683331072 |
|---|---|
| author | Chen, Siyu Liu, Kangcheng Wang, Chen Yuan, Shenghai Yang, Jianfei Xie, Lihua |
| author2 | School of Electrical and Electronic Engineering |
| author_facet | School of Electrical and Electronic Engineering Chen, Siyu Liu, Kangcheng Wang, Chen Yuan, Shenghai Yang, Jianfei Xie, Lihua |
| author_sort | Chen, Siyu |
| collection | NTU |
| description | Visual Odometry (VO) is vital for the navigation of autonomous systems,
providing accurate position and orientation estimates at reasonable costs.
While traditional VO methods excel in some conditions, they struggle with
challenges like variable lighting and motion blur. Deep learning-based VO,
though more adaptable, can face generalization problems in new environments.
Addressing these drawbacks, this paper presents a novel hybrid visual odometry
(VO) framework that leverages pose-only supervision, offering a balanced
solution between robustness and the need for extensive labeling. We propose two
cost-effective and innovative designs: a self-supervised homographic
pre-training for enhancing optical flow learning from pose-only labels and a
random patch-based salient point detection strategy for more accurate optical
flow patch extraction. These designs eliminate the need for dense optical flow
labels for training and significantly improve the generalization capability of
the system in diverse and challenging environments. Our pose-only supervised
method achieves competitive performance on standard datasets and greater
robustness and generalization ability in extreme and unseen scenarios, even
compared to dense optical flow-supervised state-of-the-art methods. |
| first_indexed | 2024-10-01T03:53:57Z |
| format | Journal Article |
| id | ntu-10356/178995 |
| institution | Nanyang Technological University |
| language | English |
| last_indexed | 2024-10-01T03:53:57Z |
| publishDate | 2024 |
| record_format | dspace |
| spelling | ntu-10356/1789952024-07-19T15:39:41Z Salient sparse visual odometry with pose-only supervision Chen, Siyu Liu, Kangcheng Wang, Chen Yuan, Shenghai Yang, Jianfei Xie, Lihua School of Electrical and Electronic Engineering Engineering Deep learning Optical flow Visual Odometry (VO) is vital for the navigation of autonomous systems, providing accurate position and orientation estimates at reasonable costs. While traditional VO methods excel in some conditions, they struggle with challenges like variable lighting and motion blur. Deep learning-based VO, though more adaptable, can face generalization problems in new environments. Addressing these drawbacks, this paper presents a novel hybrid visual odometry (VO) framework that leverages pose-only supervision, offering a balanced solution between robustness and the need for extensive labeling. We propose two cost-effective and innovative designs: a self-supervised homographic pre-training for enhancing optical flow learning from pose-only labels and a random patch-based salient point detection strategy for more accurate optical flow patch extraction. These designs eliminate the need for dense optical flow labels for training and significantly improve the generalization capability of the system in diverse and challenging environments. Our pose-only supervised method achieves competitive performance on standard datasets and greater robustness and generalization ability in extreme and unseen scenarios, even compared to dense optical flow-supervised state-of-the-art methods. National Research Foundation (NRF) Submitted/Accepted version This work was supported by National Research Foundation, Singapore through its Medium Sized Center for Advanced Robotics Technology Innovation. 2024-07-15T08:23:17Z 2024-07-15T08:23:17Z 2024 Journal Article Chen, S., Liu, K., Wang, C., Yuan, S., Yang, J. & Xie, L. (2024). Salient sparse visual odometry with pose-only supervision. IEEE Robotics and Automation Letters, 9(5), 4774-4781. https://dx.doi.org/10.1109/LRA.2024.3384757 2377-3766 https://hdl.handle.net/10356/178995 10.1109/LRA.2024.3384757 2-s2.0-85189782866 http://arxiv.org/abs/2404.04677v1 5 9 4774 4781 en IEEE Robotics and Automation Letters © 2024 IEEE. All rights reserved. This article may be downloaded for personal use only. Any other use requires prior permission of the copyright holder. The Version of Record is available online at http://doi.org/10.1109/LRA.2024.3384757. application/pdf |
| spellingShingle | Engineering Deep learning Optical flow Chen, Siyu Liu, Kangcheng Wang, Chen Yuan, Shenghai Yang, Jianfei Xie, Lihua Salient sparse visual odometry with pose-only supervision |
| title | Salient sparse visual odometry with pose-only supervision |
| title_full | Salient sparse visual odometry with pose-only supervision |
| title_fullStr | Salient sparse visual odometry with pose-only supervision |
| title_full_unstemmed | Salient sparse visual odometry with pose-only supervision |
| title_short | Salient sparse visual odometry with pose-only supervision |
| title_sort | salient sparse visual odometry with pose only supervision |
| topic | Engineering Deep learning Optical flow |
| url | https://hdl.handle.net/10356/178995 http://arxiv.org/abs/2404.04677v1 |
| work_keys_str_mv | AT chensiyu salientsparsevisualodometrywithposeonlysupervision AT liukangcheng salientsparsevisualodometrywithposeonlysupervision AT wangchen salientsparsevisualodometrywithposeonlysupervision AT yuanshenghai salientsparsevisualodometrywithposeonlysupervision AT yangjianfei salientsparsevisualodometrywithposeonlysupervision AT xielihua salientsparsevisualodometrywithposeonlysupervision |