ORB Keypoints Minimum distance?
Are there a similar min.distance values like good features to track in the ORB algorithm? https://docs.opencv.org/3.4/d4/d8c/tu...
I was doing some matching but I wanted to have unique keypoints. ORB produced duplicate or very close overlapping key-points. Or is that the limitation of ORB algorithm?
Orb does multi-scale FAST, are you sure the points in close proximity aren't at different scales? Is non-maxima suppression enabled? What is the non-maxima suppression window size? Have you looked at ANMS (adaptive non-maximal suppression) functions for even keypoint distribution?
hi @Der Luftmensch, I do not know how to implement the ANMS in my ORB algorithm
orb = cv2.ORB_create(50, 1.2, 8, 31, 0, 2, 0, 31, 20)this is my initial setup of ORBthe main issue is that the keypoints coordinates are too close to each other, different by a 1 or 2 pixels.
Do you need a multi-scale detector? Are your images at the same scale? If so, just use FAST for detection and FREAK for description. Don't implement ANMS within ORB, do it as a post-processing step to any feature detection algorithm. Search this site for ANMS pointers. You should know that ORB selects the X features (in your case, 50) that give the highest response, regardless of distribution over the image. Depending on your image this could lead to clusters of points rather than even distribution.
Sorry for the late comment, @Der Luftmensch, I would need to detect specific points for tracking and I would want the points detected to be distributed and evenly spaced out between the points instead of Cluster of Points. Or maybe if there is a cluster, I would want to remove/average them into a single point so that I would get a much more distinct keypoints to track