Rolling Shutter Homography and its Applications.

Area of Science:

• Computer Vision
• Multi-view Geometry
• Image Processing

Background:

• Homography is crucial in multi-view geometry but its adaptation to Rolling Shutter (RS) images remains underexplored.
• RS imaging introduces motion artifacts that complicate traditional geometric transformations.

Purpose of the Study:

• To adapt the concept of homography for Rolling Shutter (RS) images.
• To develop computational methods for RS Homography and apply them to key computer vision problems.

Main Methods:

• Representing point-to-point relationships in RS images using 3-8 atomic 3x3 matrices (RS Homography).
• Developing linear solvers for computing RS Homography from point correspondences.
• Deriving linear and closed-form solutions for image stitching and plane-based relative pose computation with RS images.

Main Results:

• Demonstrated that RS Homography can be expressed as a set of atomic matrices.
• Proposed effective linear solvers for RS Homography computation.
• Achieved superior performance compared to state-of-the-art methods on synthetic and real RS image datasets.

Conclusions:

• The proposed RS Homography framework effectively extends homography to RS imaging.
• The developed methods provide robust solutions for image stitching and relative pose computation with RS images.
• This work offers significant advancements for handling geometric transformations in RS imagery.