Line Features and Hybrid Methods
Line features can complement feature points in most computer vision applications and bring additional robustness.
Why it is Important
- Lines are widespread in human-made environnements, including in texture-less areas where points are scarce.
- They have a longer extent in images and can thus provide additional constraints compared to points.
- Lines provide structural and 3D information, even from a single image.
Line Detection: Line segments need a robust and accurate detector to be able to be used in generic applications.
Line Matching: After detection, lines can be matched across views to give access to 3D information.
Applications: Line features are widely used in many computer vision tasks, such as:
- Vanishing point estimation
- 3D reconstruction
- Visual Localization
Line features are promising features that should be used in combination with feature points. They complement well the latter in indoor scenarios and texture-less areas, and can provide additional accuracy and robustness to standard local feature pipelines.
- Handbook on Leveraging Lines for Two-View Relative Pose Estimation (3DV 2024) [Paper]
- GlueStick: Robust Image Matching by Sticking Points and Lines Together (ICCV 2023) [Project page]
- Vanishing Point Estimation in Uncalibrated Images with Prior Gravity Direction (ICCV 2023) [Project page]
- 3D Line Mapping Revisited (CVPR 2023) [Project page]
- DeepLSD: Line Segment Detection and Refinement with Deep Image Gradients (CVPR 2023) [Project page]
- SOLD²: Self-supervised Occlusion-aware Line Description and Detection (CVPR 2021) [Project page]