Region-based automatic mapping of tsunami-damaged buildings using multi-temporal aerial images

Abstract

After a disaster, prompt distribution of information is critical for national or local governments to plan the disaster response and recovery measures. In case of a tsunami, information about buildings destroyed by the waves is required. Here, we present a method that identifies individual damaged buildings by using aerial images obtained pre- and post-tsunami. The method utilizes significant height changes in building regions to assess the damage. Stereo aerial images are used to generate a digital surface model (DSM) of the area. We assume two cases: if geographic information system (GIS) data (building region data) are available, we use them and if GIS data are unavailable, we instead use segmented results and a filtered DSM. In each case, regions corresponding to buildings are identified in the pre-tsunami image. Damaged regions are then extracted by considering the height change within a building region between the pre- and post-disaster images. Horizontal shifts resulting from land deformation caused by the earthquake are automatically estimated by an existing algorithm such as scale-invariant feature transform (Lowe in Int J Comput Vis, 60(2):91–110, 2004). Validation showed that the proposed method extracted damaged buildings with high accuracy (94–96 % in number and 96–98 % in area) when GIS data are available and with lower accuracy (69–79 % in area) when GIS data are unavailable. In addition, we found that horizontal shifts between pre- and post-disaster should be considered to extract the damaged buildings. We conclude that our method can automatically generate effective maps of buildings damaged not only by tsunamis but also by other disasters.