Author: Kang
Year: 2022
- equirectangular rectification -> low performances for stereo matching due to severe distortion around the poles
- novel subdivision scheme with spherical geodesic grid: one to one mapping from 2D images to spherical coordinates
- Latitude / Longitude = equirectangular projection (using angles)
- cubeMap projection reduces distortion but cannot be used for stereo matching due to complex computation for referencing
- geodesic grid: projection using a dome composed of planar triangles
- No mathematicall relation between 3D image and 2D plane with a geodesic grid
- uses a icosahedral geodesic polyhedron as a geodesic grid (20 faces, 30 edges, and 12 vertices)
- when
$\pi$ and$\pi^{-1}$ are not defined mathematically, we need a look up table that is computationnally expensive especially many forward and backward projection are needed for stereo matching - the principal polyhedral triangles can be subdivided to refine the resolution of the grid
- subdivision of the polyhedron that divides epiline equally
- epipolar lines are represented as three connected straight lines on the 2D image
- stereo matching simplified with hexagonal shaped matching block