Large Scale Human Mobility Prediction in Real-time using High Performance Particle Filter

Abstract

Real-time estimation of human mobility after a massive disaster plays a crucial role in disaster relief. Traffic condition estimation with mobile phone data using data assimilation techniques recently becomes attractive research field. One of the researches studies the real-time estimation of human mobility following a massive disaster using data assimilation techniques. However, the target area of the previous study is narrower area than the inside of the Yamanote line. Therefore, we aim to estimate the real-time human mobility following a massive disaster in the area covering all of the Kanto region. In order to reduce the process time, we employ the fast algorithm to calculate the earth mover's distance (EMD). Moreover, the result of EMD is converted to a bipartite graph. In the experiment, we evaluate the computational cost and accuracy using real GPS data.