I am using features2d modules for object recognition.
Sometime it can happen that an object with many kepoints (for example a tree) will get matched to a completely different trained object.
How could I check that the keypoints from the tree are positioned gemetrically differnet than those on the object?
Please note I can't use the homography, because I have multiple training images of the same object (I am using hough voting not ransac+homography).
One solution would be that after we identified our candidate object we simply check for example its histogram versus the mean histogram of my object. But I would like to use something that verifies the gemeotric correspondecnes between points.