Recognizing Buildings through Deep Learning: A Case Study on Half-timbered Framed Buildings in Calw City

Konstantinos Makantasis, Nikolaos D. Doulamis, Athanasios Voulodimos

Abstract: Automatic detection and recognition of specific types of urban buildings is extremely important for a variety of applications ranging from outdoor urban reconstruction to navigation. In this paper we propose a system for the automatic detection and recognition of urban buildings. Most of the existing work relies on the exploitation of handcrafted features for recognizing buildings. However, due to their complex structure it is rarely a priori known which features are important for the recognition task. Our method overcomes this drawback by exploiting a deep learning framework, based on convolutional neural networks, which automatically construct highly descriptive features directly from raw data. We evaluate the performance of our method on the recognition of half-timbered framed buildings in Calw city in Germany.