Author: Kitamura
Year: 2015
- detect corner in discrete spherical images
- longitude, lattitude leads to strong distorsion
- sphere representation = isotropic ie the shape of an object is independant of the cam orientation
- FAST detector never detects at image boundaries
- tessalation base is an icosahedron
- SIF image for processing : 2D representation of 3D shape without projection and interpolation (Spherical image format - Li)
- SHarris is slow due to spherical gradient computation
- Uses an hexagonal region around the pixel to compute the score
- Uses only 4 pixels to vote for a non corner to speed up the process
- look up table to save relationship between SIF and discrete spherical image for every pixels
- able to detect very distorted corners
- but way slower than FAST (maybe due to opencv implementation)