FAST corner detection

The FAST corner detection uses simple pixel intensity comparisions to detect corners.

  1. A pixel is selected and a threshold is chosen.
  2. With the pixel as center, a set of 16 pixels on a circle of radius 3 around it are chosen.
  3. Now, for some (which is commonly 12), if there are consecutive points whose brightness is outside the range (pixel intensity threshold); then the pixel is defined as a corner.
    Note

    Other commonly seen values include 9, 11. The detector is referred to as FAST-9, FAST-11 and so on.

Preprocessing

To further optimize FAST-12, the pixels at indices - 0, 4, 8, 12 can be checked first to see if they fall outside the threshold. If they do not, then we can just exclude the point as a corner.

Postprocessing

Since, FAST will register corners on closeby pixels, non maximal suppression has to be done on the final result to combine the closer corners into one.