Image‐based detection of structural elements in 2D steel hall floor plans for automated BIM modelling

Abstract

This paper presents an approach for the detection of structural elements in 2D steel hall floor plans using deep learning techniques. The proposed framework employs YOLO‐based Convolutional Neural Networks to identify key components such as columns, structural grids, and foundation footings in engineering drawings. The methodology is based on automatically generated synthetic training data derived from parametric BIM models, enabling precise annotation without manual labelling. The study investigates the influence of input image resolution and model capacity on detection performance, with particular emphasis on small‐object detection in densely annotated drawings. The results demonstrate reliable identification of structural elements and sufficient geometric accuracy for further engineering interpretation, while highlighting the importance of input image resolution for maintaining detection performance. A limitation of the study is that the dataset consists exclusively of synthetically generated drawings, without validation on real‐world scanned documentation. Although the current work is limited to 2D detection, the proposed framework can serve as a basis for future integration with BIM environments.