Autonomous Vehicle/Video Geometry

FAST (Features from Accelerated Segment Test)

Naranjito 2025. 4. 3. 19:31
  • It's a fast corner detection method often used in real-time applications due to its speed and simplicity.

 

 

To determine if a pixel p is a corner, FAST uses:

  • A circle of 16 pixels around the candidate pixel p.
  • These pixels are tested to see if a segment of N contiguous pixels on the circle are all brighter or all darker than p by a threshold.

 

  • Select 4 key pixels around p (in this example: pixels 1, 5, 9, 13).
  • Check any 3 out of these 4 first:
    • If not enough contrast, skip (early rejection).
    • If contrast is good, check the full circle of 16 pixels.
  • If at least 12 contiguous pixels are all brighter or all darker than p, then:
    • Pixel p is classified as a corner.

  •  Why is this fast?
  • Early rejection: Only a few checks are needed in most cases.
  • It reduces computation by quickly rejecting non-corners before checking all 16 pixels.

'Autonomous Vehicle > Video Geometry' 카테고리의 다른 글

Homography  (0) 2025.04.04
BRIEF (Binary Robust Independent Elementary Features)  (0) 2025.04.03
KAZE  (0) 2025.04.03
SURF (Speeded Up Robust Features)  (0) 2025.04.03
SIFT (Scale-Invariant Feature Transform)  (0) 2025.04.03