Deep learning-based method for reconstructing three-dimensional building cadastre models from aerial images

Nowadays, many of the world’s large cities are faced with the issue of land scarcity for construction due to the increasing growth of urbanization, as well as the economic downturn for exploiting lands and properties, and city officials have come up with the idea of optimal management of real estate in order to cope with these problems. The purpose of our study is to reconstruct three-dimensional (3-D) building cadastre models (3DBCMs) with an approach to improve the state of land administration in Tehran metropolis. Our study is being implemented and evaluated in three stages. The first stage involves collecting aerial images. The interior and exterior orientation parameters are preprepared in this step. The second stage involves automatic interpretation and extraction of buildings from aerial images by providing a method of interpretation called fully automatic interpretation with deep learning (FAIDL). The third stage involves 3-D building modeling and evaluating the effect of FAIDL method on the automatic interpretation of images. The results showed that the 3-D models of building have a better geometric accuracy compare to 60 cm, which are produced with the proposed algorithm.