Automated building geometry modeling using UAV imagery and deep learning for energy simulation

초록

Urban building energy modeling (UBEM) plays a critical role in evaluating and optimizing energy performance across city-scale building stocks. However, acquiring high-quality geometric data remains time-consuming and challenging, leading many UBEM approaches to rely on simplified building models that introduce significant inaccuracies in energy simulation. This study proposes an automated framework for extracting key geometric features including building height, number of floors, and window-to-wall ratios (WWRs) by integrating unmanned aerial vehicles (UAVs), LiDAR point clouds, and YOLOv5-based deep learning object detection. The approach significantly reduces manual modeling effort and enables the generation of detailed building models with minimal user input. Case studies on multiple institutional buildings demonstrate that incorporating accurate window geometry results in up to 14.2 % deviation in heating energy demand compared to simplified models. Although the current validation focuses on the building scale, the methodology is designed to be scalable and adaptable to larger groups of buildings or future urban-scale applications. The findings underscore the importance of fa & ccedil;ade-level geometric detail in energy simulation accuracy and demonstrate the potential of UAV-and AI-based automation for advancing data-driven building energy modeling practices.

키워드