Panoramic Stitching of Cylindrical Surfaces
Hokkanen, Leevi (2024)
2024
Tietotekniikan DI-ohjelma - Master's Programme in Information Technology
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2024-05-31
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202405276346
https://urn.fi/URN:NBN:fi:tuni-202405276346
Tiivistelmä
Many of the modern stitching algorithms assume the pictured surface to be planar-shaped and the camera-motion be purely rotational leading into issues with perspective alignment of the out put panoramic images. This can cause issues in panoramic imaging that is done in contexts like civil engineering or medical imaging, where accurate picturing is necessary for tasks like quality assurance.
This paper discusses the use of structure-from-motion analysis to estimate cylindrical warping parameters and their use on picturing the surface of cylindrical tunnels with the goal of mitigating perspective misalignments. Structure-from-Motion refers to the act of recovering the scene 3D structure and camera parameters from images while cylindrical warping refers to the act of pan ning image coordinates to a cylindrical shape.
The cylindrical surfaces are captured systematically along the tunnel surface, from which the sur face features and homography estimations are handled via SIFT features. The scene pose-re covery and 3D scene reconstruction are achieved via essential and fundamental matrix estima tion, and triangulation. These features can be utilized for obtaining cylindrical warping parameters like the camera orientation and cylinder radius, which allows for shape and orientation aware warping of images. The final stitching to the warped images is handled via known techniques like OpenCV stitcher-class.
Results show improved perspective alignment with cylindrical warping, although challenges such as parameter tuning, resource consumption, and artifacts noticeable with the results. Other criti cism includes reliance on manual parameter assignment, and the need for significant image over lap. Despite these limitations, the proposed method offers promising results in aligning tunnel images, with opportunities for further optimization and refinement.
This paper discusses the use of structure-from-motion analysis to estimate cylindrical warping parameters and their use on picturing the surface of cylindrical tunnels with the goal of mitigating perspective misalignments. Structure-from-Motion refers to the act of recovering the scene 3D structure and camera parameters from images while cylindrical warping refers to the act of pan ning image coordinates to a cylindrical shape.
The cylindrical surfaces are captured systematically along the tunnel surface, from which the sur face features and homography estimations are handled via SIFT features. The scene pose-re covery and 3D scene reconstruction are achieved via essential and fundamental matrix estima tion, and triangulation. These features can be utilized for obtaining cylindrical warping parameters like the camera orientation and cylinder radius, which allows for shape and orientation aware warping of images. The final stitching to the warped images is handled via known techniques like OpenCV stitcher-class.
Results show improved perspective alignment with cylindrical warping, although challenges such as parameter tuning, resource consumption, and artifacts noticeable with the results. Other criti cism includes reliance on manual parameter assignment, and the need for significant image over lap. Despite these limitations, the proposed method offers promising results in aligning tunnel images, with opportunities for further optimization and refinement.