An Approach to Automatic Detection of Architectural Façades in Spherical Images

The fast and automatic facade extraction method is an important step to retrieve actual information about cities. It helps to maintain a public registry, may be used to examine the technical condition of buildings or in process of city planning. We propose an automatic method to detect and retrieve building facades from spherical images. Our method uses deep learning models trained on automatically labelled spherical images collected in the virtual city we have generated. Finally, we compare the proposed solution using three different deep learning models with a classic method. Our experiment revealed that the proposed approach has a similar or better performance than the current methods. Moreover, our solution works with unprepared data, while existing methods require data pre-processing.