AN INNOVATIVE APPROACH FOR THE SEMANTIC SEGMENTATION OF SURVEYED BUILDING FACADES LEVERAGING ON ARCHITECTURAL DRAWINGS

Traditionally, drawing products created from 3D surveying activities have been the universal medium of communication used by architects. This has resulted in a vast repository of graphic documentation that serves as a testament of the architectural heritage. The embedded information found in elevations, plans and sections holds considerable value, and it can be seamlessly integrated into the intricate graphics produced during large-scale data acquisition processes. The core objective of this research is to investigate how the information coming from the large amount of existing architectural technical drawings can support 3D heritage classification processes and avoid time-consuming annotation of materials and construction techniques of historical building facades. Starting from available sets of drawings, AI-based methodologies are applied for the annotation of orthoimages and point clouds in order to obtain a predictive model that can recognize classes of materials and construction techniques in a large amount of data. The predicted classes also allow the automatic creation of vector drawing representing the facades of new buildings, providing a novel tool to facilitate the processes of analysis and conservation of architectural heritage.