There are several photos of the same object, but from different angles, of different size and quality. It is necessary to programmatically determine the key points on these photos, and without any templates.
In this case,
SIFT / SURFhas not yet come up with anything because of the invariance of these keypoints with respect to many types of transformations. Implementations are present in libraries like
Orb  , which is also implemented in
OpenCV, see http://opencv.itseez.com/modules/features2d/doc/feature_detection_and_description.html?highlight=orb#ORB
Something tells me that the task can be reduced to finding the eigenvectors of the matrices obtained from the photos. But I'm not sure, unfortunately.