Deep learning-based text detection on architectural floor plan images

Abstract Architectural drawings are an important source of information for many construction-related tasks, as they contain geometric and semantic about building parts. However, the manual extraction room stamps insertion such gleaned into facility management systems is quite laborious and, thus, its automation anticipated. In this paper, a method proposed to detect classify obscure or illegible text elements on legacy 2D architectural possibly poor quality. contrast existing approaches, deep learning model specifically trained task at hand rather than making use transfer approaches. The resulting snippets can be further processed with natural language processing tools fed system automatically. Other conceivable applications include drawing header information, material type any additional given drawing, facilitate enrichment digital twins structures data. To provide training data, two floor plan datasets annotated in consistent manner. influence different data augmentation techniques investigated systematically. With regard performance efficiency, presented compared alternative hand.