Building Damage Assessment Using AI and Video from Vehicle-Mounted Camera and Damage Rate Evaluation by Damage Map

Abstract

Japan frequently experiences major earthquakes. Damage assessment of buildings, such as the Emergency Safety Check, is required for reconstruction efforts. Currently, damage assessment is performed by investigators through field surveys. In this study, the authors attempted to propose the assessment using AI and IoT for more effective and faster assessment. First, the deep learning model was constructed using the images of damaged buildings from previous earthquakes. Subsequently, the authors created a damage map by using the results of the damage assessment of buildings by the learning model and the aerial images taken by a drone in Suzu, Ishikawa; and then, the authors compared it with the visual assessment. The results showed that the model was able to adequately determine the degree of total destruction, however the model tended to underestimate the degree of semi-destruction.