Digital reconstruction of District Six architecture from archival photographs

Abstract

In this thesis we present a strategy for reconstructing instances of District Six Architecture from small sets of old. uncalibrated photographs that are located in the District Six Museum photographic archive. Our reconstruction strategy comprises two major parts. First, we implement a geometry reconstruction framework. based on work by Debevec et al. [1996]. This is used to reconstruct the geometry of a building given as little input as a single photograph. The approach used in this framework requires the user to design a basic model representing the building at hand. using a set of geometric primitives, and then define correspondences between the edges of this model and the edges of the building that are visible in the photographs. This approach is effective, as constraints inherent III the geometry of architectural scenes are exploited through the use of these primitives. The second component of the reconstruction strategy involves texturing the reconstructed models. To accomplish this, we use a combination of the original textures extracted from the photographs, and synthesized textures generated from samples of the original textures. For each face of the reconstructed model, the user is able to use either the original texture material. synthesized material, or a combination of both to create desirable results. Finally, to illustrate the effectiveness of our reconstruction strategy, we consider three example cases of District Six architecture and their reconstructions. All three examples were reconstructed successfully, and using findings from these results, critical analyses of both aspects of our strategy are presented.