Classification of structural building damage grades from multi-temporal photogrammetric point clouds using a machine learning model trained on virtual laser scanning data

Automatic damage assessment by analysing UAV-derived 3D point clouds provides fast information on the situation after an earthquake. However, of different grades is challenging given variety in characteristics and limited transferability methods to other geographic regions or data sources. We present a novel change-based approach automatically assess multi-class building from real-world using machine learning model trained virtual laser scanning (VLS) data. Therein, we (1) identify object-specific cloud-based change features, (2) extract changed parts k-means clustering, (3) train random forest with VLS based (4) use classifier photogrammetric clouds. evaluate respect its capacity classify three (heavy, extreme, destruction) pre-event post-event earthquake L’Aquila (Italy). Using features derived bi-temporal clouds, our transferable multi-source input used for training application (real-world photogrammetry). further achieve simulated which characterises across regions. The yields high multi-target classification accuracies (overall accuracy: 92.0%–95.1%). Classification performance improves only slightly when region-specific (< 3% higher overall accuracies). consider especially relevant applications where timely required sufficient not available.