Automated Recognition of Building Façades for Creation of As-Is Mock-Up 3D Models

Abstract

Recent advancements in image sensors, photogrammetry, and image processing have enabled continuous documentation of buildings’ as-is conditions. Generating accurate three-dimensional (3D) models of buildings is currently considered as a time-consuming, semimanual, and expensive process. However, building models with a low level of details can be created through an automated process rather inexpensively and be used for a variety of purposes, such as virtual urban modeling, predisaster planning, and postdisaster recovery studies. This paper presents an automated method for generating mock-up 3D models of buildings using ground-based images. Rectified images of building façades along with their two-dimensional (2D) footprint were used to create 3D models of buildings with dimension errors less than 40 cm. The method uses the gradient profile of images to predict the locations of the common façade architectural elements (i.e., windows and doors). These locations are then used to extract pixel- and texton-level local and global features to classify the façade elements. The layout of the elements is used as an input to develop a split grammar for the façade. Ultimately, the grammar is used to reconstruct the structure of the façade and to generate the mock-up model. A public façade image database as well as three case study buildings are used to evaluate the performance of the proposed method. The results indicated an overall average accuracy of 80.48% in classifying the architectural elements. The proposed method is neither dependent on large-scale labeled ground-truth data nor requires hours for training classifiers. In addition, it has the potential to be scaled up and applied to buildings with a repetitive façade structure.