Once disaster occurs, we have to understand the whole picture of damage situation. However, there is not enough human resources and time to do it. In 2019, we were affected by Yamagata earthquake, and many buildings were damaged in Murakami city. Most of buildings damage were concentrated on their roofs. Local responders tried to inspect those roof damage, however they cannot do it from ground. Against this issue, we decided to take images of roof damage utilizing drones. We designed the flight plan covering over affected area, operated a drone, and got images. Aftermath, we created orthophoto mosaic from those images. We published it for local responders to inspect roof damage of each building in the web-based GIS platform. Furthermore, we detect roof damage by human and let AI learn the result of our detection of roof damage. This is followed the framework of Human-in-the-Loop. Just now, accuracy of the roof-damage detection by AI was not so high. In this paper, we introduce the work-flow of this challenge from designing flight plan for drone to roof-damage detection by AI which is educated with images of actual damage situation after disaster occurrence.
展开▼
