Building Facade Sketch Drawing Based on Computer Vision

Buildings are widely known as the most crucial element of urban landscapes, and extracting building elevations is a cornerstone of many modeling procedures. Although laser scanning can accurately obtain building façade information, its use is too expensive. This paper introduces a novel approach to automatically generate sketches of building facades using computer vision technology when building facade images are captured. We employed computer vision algorithms with CAD secondary development technology to extract the facade and windows from a single building image and convert them into a simplified CAD style. Initially, the proposed method conducts semantic segmentation to distinguish various regions of the building. Next, the 2D coordinates of the identified objects are automatically extracted, and the extracted 2D point set coordinates are fitted and inputted into the CAD system for automatic drawing. The generated sketch effectively captures the building’s facade information while also accurately representing the dimensions of the windows and facade. Our proposed method has undergone evaluation using real-world building image data. The experimental results demonstrate that the proposed approach can generate precise facade sketches and attain 85% recognition accuracy for windows. The entire process, starting from semantic segmentation to mapping CAD drawings, can be completed in 2–2.5 s. This research has potential applications in applying computer vision techniques in modeling in architecture, urban planning, and computer graphics.