PLL-VO： An Efficient and Robust Visual Odometry Integrating Point-Line Features and Neural Networks
Visual odometry is crucial for the navigation and planning of autonomous robots, but low-light conditions, dramatic lighting changes, and low-texture scenes pose significant challenges to odometry estimation. This paper proposes PLL-VO, which integrates point-line features and deep learning. To over...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Copernicus Publications
2025-07-01
|
| Series: | ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences |
| Online Access: | https://isprs-annals.copernicus.org/articles/X-G-2025/1045/2025/isprs-annals-X-G-2025-1045-2025.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1850112485135220736 |
|---|---|
| author | L. Zhao Y. Yang D. Ma X. Lin W. Wang |
| author_facet | L. Zhao Y. Yang D. Ma X. Lin W. Wang |
| author_sort | L. Zhao |
| collection | DOAJ |
| description | Visual odometry is crucial for the navigation and planning of autonomous robots, but low-light conditions, dramatic lighting changes, and low-texture scenes pose significant challenges to odometry estimation. This paper proposes PLL-VO, which integrates point-line features and deep learning. To overcome the impact of complex lighting conditions, a self-supervised learning method for interest point detection and a line detection algorithm that combines line optical flow tracking with cross-constraints is presented. After selecting keyframes based on point feature counts and line feature overlap angles, we integrate convolutional neural networks (CNNs) and graph neural networks (GNNs) to enhance sparse matching, thereby improving both accuracy and computational efficiency. PLL-VO system are evaluated in multiple datasets under various lighting conditions, demonstrating a 6.3% reduction in absolute trajectory error for pose estimation compared to state-of-the-art (SOTA) algorithms, the average computation time for visual odometry (43 ms) shows a 29.74% decrease compared to the state-of-the-art (SOTA) algorithms. |
| format | Article |
| id | doaj-art-16fdf389c1f64f728a58ee3ad1c72bb4 |
| institution | OA Journals |
| issn | 2194-9042 2194-9050 |
| language | English |
| publishDate | 2025-07-01 |
| publisher | Copernicus Publications |
| record_format | Article |
| series | ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences |
| spelling | doaj-art-16fdf389c1f64f728a58ee3ad1c72bb42025-08-20T02:37:21ZengCopernicus PublicationsISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences2194-90422194-90502025-07-01X-G-20251045105210.5194/isprs-annals-X-G-2025-1045-2025PLL-VO： An Efficient and Robust Visual Odometry Integrating Point-Line Features and Neural NetworksL. Zhao0Y. Yang1D. Ma2X. Lin3W. Wang4School of Architecture and Urban Planning, Shenzhen University, Shenzhen 518060, ChinaSchool of Geomatics and Geographical Sciences, Liaoning Technical University, Fuxin 123000, ChinaSchool of Architecture and Urban Planning, Shenzhen University, Shenzhen 518060, ChinaSchool of Architecture and Urban Planning, Shenzhen University, Shenzhen 518060, ChinaSchool of Architecture and Urban Planning, Shenzhen University, Shenzhen 518060, ChinaVisual odometry is crucial for the navigation and planning of autonomous robots, but low-light conditions, dramatic lighting changes, and low-texture scenes pose significant challenges to odometry estimation. This paper proposes PLL-VO, which integrates point-line features and deep learning. To overcome the impact of complex lighting conditions, a self-supervised learning method for interest point detection and a line detection algorithm that combines line optical flow tracking with cross-constraints is presented. After selecting keyframes based on point feature counts and line feature overlap angles, we integrate convolutional neural networks (CNNs) and graph neural networks (GNNs) to enhance sparse matching, thereby improving both accuracy and computational efficiency. PLL-VO system are evaluated in multiple datasets under various lighting conditions, demonstrating a 6.3% reduction in absolute trajectory error for pose estimation compared to state-of-the-art (SOTA) algorithms, the average computation time for visual odometry (43 ms) shows a 29.74% decrease compared to the state-of-the-art (SOTA) algorithms.https://isprs-annals.copernicus.org/articles/X-G-2025/1045/2025/isprs-annals-X-G-2025-1045-2025.pdf |
| spellingShingle | L. Zhao Y. Yang D. Ma X. Lin W. Wang PLL-VO： An Efficient and Robust Visual Odometry Integrating Point-Line Features and Neural Networks ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences |
| title | PLL-VO： An Efficient and Robust Visual Odometry Integrating Point-Line Features and Neural Networks |
| title_full | PLL-VO： An Efficient and Robust Visual Odometry Integrating Point-Line Features and Neural Networks |
| title_fullStr | PLL-VO： An Efficient and Robust Visual Odometry Integrating Point-Line Features and Neural Networks |
| title_full_unstemmed | PLL-VO： An Efficient and Robust Visual Odometry Integrating Point-Line Features and Neural Networks |
| title_short | PLL-VO： An Efficient and Robust Visual Odometry Integrating Point-Line Features and Neural Networks |
| title_sort | pll vo an efficient and robust visual odometry integrating point line features and neural networks |
| url | https://isprs-annals.copernicus.org/articles/X-G-2025/1045/2025/isprs-annals-X-G-2025-1045-2025.pdf |
| work_keys_str_mv | AT lzhao pllvoanefficientandrobustvisualodometryintegratingpointlinefeaturesandneuralnetworks AT yyang pllvoanefficientandrobustvisualodometryintegratingpointlinefeaturesandneuralnetworks AT dma pllvoanefficientandrobustvisualodometryintegratingpointlinefeaturesandneuralnetworks AT xlin pllvoanefficientandrobustvisualodometryintegratingpointlinefeaturesandneuralnetworks AT wwang pllvoanefficientandrobustvisualodometryintegratingpointlinefeaturesandneuralnetworks |