人物移動データからの滞留点推定を目的としたクラスタリング手法と評価指標

抄録

A monitoring system for the elderly by using a Laser Range Finder (LRF) is being developed. The long-term objective of the system is to extract daily life patterns of an elderly person based on trajectory data and to detect an anomaly if, when observing a new life pattern, it significantly differs from the daily life patterns. A life pattern represents a series of behavior transitions such as Toilet => Washstand => Closet => Bed. In such places, it is assumed that an elderly person stays for a certain time of period. In this paper, such places are called “stay points”. This research focuses on those methods that automatically estimate the stay points from movement data. Clustering approach is employed for this purpose. Numerical experiments were conducted by considering situations that may occur in elderly facilities. Through a series of computational experiments, we investigate which combination of clustering method and evaluation index provides the best estimation of the stay points.