Flood Depth Assessment with Location-Based Social Network Data and Google Street View - A Case Study with Buildings as Reference Objects

Flood damage accurate assessment, such as flood depth, is very helpful for disaster relief. However, most of existing methods have some limitations, either the expensive data requirement (e.g., hydrological observations), or only producing a rough assessment with location-based social network (LBSN) data. To obtain an accurate and real-time flood damage assessment, this paper firstly collects pairs of pre- and post-flood images for the same location from LBSNs and Google Street View respectively. Next, buildings were segmented in images using machine learning methods (e.g., Mask R-CNN). Finally, the rectifying and building height extraction were developed to calculate the accurate value of the flood depth by comparing the paired images. The method was evaluated by using the real datasets generated from flood disasters in Japan and US. Experiment results show that the proposed method provides an accurate and real-time flood disaster assessment.