Using a SIFT feature detection and a FLANN feature matcher, I achieve the following result:
The homograph is drawn in white, but due to the features on the left, it's way to wide.
Now; what would be the way to filter the matching features, so only the one of the tower remain? In fact, I want to classify/group the features that are close together (assuming this means these features describe all one the same building).
I think mutual distance is the key here, as I want to group the features that are close together.. So I suppose something like SVM could work? Or are the better/easier ways to classify these points? ( as a machine learning algorithm seems a bit too match)
For sake of illustration, this is what I want to achieve: