I want to make some real time image comparison. FAST seems to perform well, but for matching purposes I have difficulties getting the keypoints described correctly.
In my testing, I tried the following code:
import cv2
# initiate some stuff
detector = cv2.FastFeatureDetector_create(20)
orb = cv2.ORB_create(150)
bf = cv2.BFMatcher()
# detect and compute keypoints for imageA in separate steps
imageA = cv2.imread("imageA.png", 0)
kpA = detector.detect(imageA, None)
kpA_computed, desA = orb.compute(imageA, kpA)
drawAkp = cv2.drawKeypoints(imageA, kpA_computed, None, flags=cv2.DRAW_MATCHES_FLAGS_DRAW_RICH_KEYPOINTS)
# detect and compute keypoints for imageB in 1 step
imageB = cv2.imread("imageB.png", 0)
kpB_computed, desB = orb.detectAndCompute(imageB, None)
drawBkp = cv2.drawKeypoints(imageB, kpB_computed, None, flags=cv2.DRAW_MATCHES_FLAGS_DRAW_RICH_KEYPOINTS)
# these shouldn't be that different from one another
cv2.imshow("A", drawAkp)
cv2.imshow("B", drawBkp)
cv2.waitKey(0)
For some reason, the keypoints in imageA are not "rich" at all.
I understand that the 2 images are using different algorithms for the keypoints detection, but I expected that imageA will have the sizes and orientations included after the compute() I think understanding the reason behind that will help me with my matching problems.