First, we detect FAST keypoints. FAST takes one parameter from the user, the threshold value between the center pixel and those in a circular ring around it. We use a ring radius of nine pixels as it gives good performance. FAST also produces keypoints that are along edges. To overcome this, we use the Harris corner measure to order the keypoints. If we want N keypoints, we keep the threshold low enough to generate more than N keypoints, and then pick the topmost N based on the Harris corner measure.
FAST does not produce multiscale features. ORB employs a scale pyramid of the image and produces FAST features (altered by Harris) at each level in the pyramid.