Unmasking Devastation: Near Real-time Earthquake-induced Building Damage Detection

The damages caused by an earthquake are devastating. Following the earthquake, supporting humanitarian activities like rescuing stranded people inside the buildings, distributing relief aid, etc. is of prime importance so as to limit the damages caused by the earthquake. In such a critical situation, state-of-the-art sensors such as LiDAR can be deployed over Unmanned Aerial Vehicle (UAV) that provides accurate depth information which can be used for detecting damaged buildings. In our work, we have proposed a building damage detection framework consisting of a deep-learning model that takes pre and post-3D point clouds as input and detects damaged buildings. Such timely information can help save the time of the rescue teams by reducing the search operations by directly accessing the locations of the damaged buildings. Our proposed approach is tested for the study area of the 2016 Kumamoto earthquake and is capable of performing change detection by detecting damaged buildings with an accuracy of 81.2% which proves the efficiency of our system. We envisage that our proposed approach can be used in disaster situations to provide the damage map of the buildings in near-real time.