Building Envelope Object Detection Using YOLO Models

Building performance significantly influences energy use and indoor thermal conditions tied to the quality of living for its occupants. Therefore, information on building envelopes is essential, especially considering that envelopes and windows can impact 50% of energy loads in the United States. However, current retrofits supporting Building Energy Modelling (BEM) tools face multiple barriers, including time consumption and labor intensity due to manual modeling and calibration processes. This paper proposes using Deep Learning (DL) -based object detection algorithms to detect building envelope components, more specificlly doors, and windows, that can be applied to building energy performance analysis, 3D modeling, and assessment of thermal irregularities. We compare four different versions of the state-of-the-art YOLO V5 model to identify which version best suits the goal of detecting these building components. Results show that YOLO V5_X provides the best performance for detection accuracy.