Development of ERESS in panic-type disasters: Disaster recognition algorithm by buffering-SVM

Many people have faced mortal risks due to sudden disasters such as fires, earthquakes, and terrorisms all over the world. To solve this kind of problems, we are developing ERESS (Emergency Rescue Evacuation Support System). This system consists of only mobile terminals with various sensors. In ERESS, these sensor data are exchanged and analyzed using only ad-hoc communication among nearby mobile terminals. ERESS aims to reduce the number of victims by supporting real-time evacuation in the critical situation for less than 30 seconds immediately after disaster outbreaks. The conventional ERESS recognized the disaster outbreaks by behavior analysis of terminal holders from sensor data using SVM (Support Vector Machine). However, there was a problem that it is difficult to obtain high reliability of judging disaster outbreak. In this paper, we propose a new method by buffering judgment results of SVM. We carry out experiments on the emergency evacuation with more than 200 examinees. We investigate the judgment accuracy of the proposed method and the time required for recognition of disaster outbreak.