Building Information Modeling and Classification by Visual Learning At A City Scale

In this paper, we provide two case studies to demonstrate how artificial intelligence can empower civil engineering. In the first case, a machine learning-assisted framework, BRAILS, is proposed for city-scale building information modeling. Building information modeling (BIM) is an efficient way of describing buildings, which is essential to architecture, engineering, and construction. Our proposed framework employs deep learning technique to extract visual information of buildings from satellite/street view images. Further, a novel machine learning (ML)-based statistical tool, SURF, is proposed to discover the spatial patterns in building metadata. The second case focuses on the task of soft-story building classification. Soft-story buildings are a type of buildings prone to collapse during a moderate or severe earthquake. Hence, identifying and retrofitting such buildings is vital in the current earthquake preparedness efforts. For this task, we propose an automated deep learning-based procedure for identifying soft-story buildings from street view images at a regional scale. We also create a large-scale building image database and a semi-automated image labeling approach that effectively annotates new database entries. Through extensive computational experiments, we demonstrate the effectiveness of the proposed method.