Generating Disaster Damage Assessment Datasets with Physics-Based Simulations and Sensing Data

This paper introduces an approach for creating a synthetic and simulation-based dataset to support unmanned aerial vehicle (UAV)-based post-disaster damage assessment. Our approach leverages physics-based simulation (e.g., applied element method) models to generate wide spectra of damaged infrastructure that compose the post-disaster scenarios, while professional engineers manually assign damage states that closely match quantitative descriptions. Through the proposed approach, each damage instance in the dataset will include essential attributes for damage evaluation, such as the disaster-induced damage state and infrastructure type. The expected dataset will provide both reality-capture visuals and damage-related information to support the training and validation of machine learning (ML) models, which in turn will enhance the accuracy and speed of large-scale post-disaster damage assessments.