FAST or ORB feature detection pattern to scan the image
Hi there,
Does ORB or FAST feature detector follow a random scanning pattern to search for features in the image, or does it scan patch by patch in a symmetrical shape the image for features? This is the step right before harris corner is applied to filter out falsely-detected corners?
Thanks!
The papers for each are easy to find and surprisingly easy to understand. You may find it valuable to actually read the original papers.
The original papers (https://www.edwardrosten.com/work/ros...) and (http://www.willowgarage.com/sites/def...) don't outline what pattern is used to select the patches. It is not a random patch as I see the same pattern of features always getting selected but it's also not clear which template is used to scan the image.
The FAST paper is only about which pattern is used to identify keypoints.
@Der Luftmensch would be great if you could point me to where I can learn about the pattern of ORB and FAST, or just tell me what is the pattern. I don't see any material that explains the exact pattern of the ORB and FAST feature detection. Thanks in advance.
Take a look at Figure 1 in the FAST paper. For all pixels in the image, a test is performed looking a ring of 16 neighboring pixels. If 12 (or 9) of the neighbors on the ring are all above or below the central pixel by some threshold, then the central pixel is considered a keypoint feature. FAST is only keypoint detection. ORB is both detection and description - Oriented FAST and Rotated Brief. The papers really are about as accessible as it gets, you should read them if you haven't yet done so.