AI: Architectural Intelligence. Deep learning and heritage environments

An unprecedented computing power and the vast datasets recently available have boosted machine learning techniques, featuring algorithms that can “learn” from experience. These are gaining importance in many disciplinary fields, dealing with computer vision and object recognition problems. Nevertheless, the applications of such tools in the fields of architecture and cultural heritage are still limited, as well as the spread of public datasets to encourage experiments and research. This contribution presents a project on artificial intelligence (AI) technologies aimed to support the management and accessibility of both architectural heritage and the connected multimedia information. Original developments of a deep learning model based on convolutional neural networks are employed in a smartphone app to recognize monuments framed using the camera. AI recognition works from different viewpoints and does not need any reference image stored on the device. Once the monument is identified, the app provides the user with information previously downloaded from the web. The Central Archaeological Area in Rome has been a test field for the first experiments. The app prototype shows that AI offers a chance to enhance existing digital archives and deal with open issues in managing spatial information.