3D Map Computation from Historical Stereo Photographs of Florence

Abstract

The analysis of early photographic sources is fundamental for documenting and understanding the evolution of a city so rich in history and art as Florence. Indeed, by the 1860s several photographers used to work in town, and their images (often obtained through stereoscopic set-ups) can help us to reconstruct Florence in 3D as it was by the time of the Italian unification. The first and most delicate part of such reconstruction process is the computation of disparity maps from the historical stereo pairs. This is a very challenging task for fully-automatic computer vision algorithms, since XIX century photographs are affected by several problems—ranging from superficial damages to asynchronous acquisition—that are usually absent in modern images. In this paper we describe a semi-automatic method that, through minimal user input, allows the creation of coherent and realistic 3D maps of florentine scenes.