Author: Shigang Li
Year: 2008
- uses spherical camera model
- conversion to longitude/lattitude images to get epipolar lines
- In the case of a fisheye image depth (z coordinate) != distance (rho in spherical coords)
- The disparity is the difference between length of arcs d = f(theta_l - theta_r)
- Images need to be rectified: a rotation matrix is computed thanks to a calibration process that can map the pixels of a camera so that the frames of both images are aligned
- High computational cost of searching along epipolar curves => creation of a latitude longitude image
- Simulation to study the error factors for distance computation: blind spot around the epipoles
Explication assez claire de la stéréo pour les fisheye, papier important