UAV-based 3D segmentation and structural damage assessment of regional buildings

Abstract

Accurate and efficient structural damage assessment is vital for disaster response and recovery. Traditional on-site methods often lack speed, safety, and precision. The emergence of computer vision techniques has revolutionized structural engineering, offering reliable alternatives to conventional methods. However, large-scale regional buildings are rarely studied. Leveraging advancements in unmanned aerial vehicle (UAV) photogrammetry and 3D modeling, this study proposes an integrated framework for aerial reconstruction, building segmentation, and damage state evaluation of earthquake-affected regional buildings. Three point-cloud modeling tools, Pix4Dmapper, Agisoft Metashape, and Bentley ContextCapture, were evaluated for their performance in structural engineering applications. A novel hybrid building segmentation approach, combining Random Sample Consensus (RANSAC) and Density-Based Spatial Clustering of Applications with Noise (DBSCAN) with grid filtering, was developed and paired with an advanced algorithm for structural damage evaluation. Residual drift ratios were used to quantify the structural damage states in alignment with the Federal Emergency Management Agency (FEMA) P-58 guidelines. Validation of the proposed framework on regional buildings damaged by the 2022 Luding earthquake revealed that Bentley ContextCapture generated the densest and most precise models, while the proposed segmentation method achieved superior accuracy compared to standard procedures. Furthermore, the successful identification of building damage states enabled rapid and precise damage evaluations, which ultimately aid disaster recovery planning. This research also provides valuable insights into the applicability of various software for different situations and illustrates the effects of critical parameters on segmentation and damage evaluation accuracy.