I have ~125 analog aerial photographs from 1937 (examples below) and need to reconstruct their locations where the images were taken for a research project. Unfortunately, little is known about them: neither if they are all taken at the same altitude nor if the images are oriented to the north. They could have been taken anywhere within the Germany's borders at the time. And the images were printed in different sizes and thus differ in print/scan resolution. Image recognition won't work because the spatial structures have changed too much since then, but I would imagine that a large part of the land parcels and streets / squares have remained the same. I found this paper by Dragos Costea and Marius Leordeanu describing "Aerial image geolocalization from recognition and matching of roads and intersections", and so far this looks to be the most promising approach for my problem, as it doesn't rely on image recognition. It would be a success for me if I can locate about 7-10% of the images.
However, since I am not a specialist and the implementation / testing means a lot of effort for me, I wonder if you think the project is realistic in principle or even know an alternative method that would be worth trying?
Update: With the help of a known subject (Monument to the Battle of the Nations) I can know determine that the images are not oriented to the north.