I’m curious, what’s the smallest set of pixels of a future identified with ORB/sift?
What’s a future anyway? Is it a square piece of an image?
1 Like
These helped me as primers:
SIFT:
ORB:
As for size… Do you mean image-scale (number of pixels, percentage of total pixels in image) or life-scale?
Because I read a number of papers where they use ORB to identify different types of unknown cells from microscope slides to compare them to known cells… So pretty freaking small!
No I mean the size of the sample from an image that’s used for matching. I believe min-num-futures is about how many samples is created to use for matching.
I can’t speak for what happens when extracting features from drone images, but in a previous job searching for Near-Earth Asteroids and Comets, a routine selected stars by determining centroids of groups of pixels x number of standard deviations above the background noise of the CCD detector. Stars were matched to star catalogues with extremely precise and (usually, for stars that hadn’t moved too much since the catalogue was created) accurate celestial positions. Groups of pixels that had reasonably consistent motion between exposures (usually 4 exposures over 30-60 mins for each field) were listed, which we manually checked to see if they were real.
I can’t recall if the box of pixels used for determining the centroid was fixed or varied, depending on how good the seeing was though (how much the stars twinkled - more = more blurry through a telescope). However, a 5X5 or 6X6 pixel box would generally be used.
3 Likes