Evolution strategy for anomaly detection in daily life monitoring of elderly people

Abstract

Recently, various types of daily life monitoring methods have been proposed for elderly care. We have proposed the concept of informationally structured space (ISS) and applied ISS using robot partners and sensor network devices to daily life monitoring. One of the most important roles in daily life monitoring is anomaly detection. Anomaly detection is to identify or detect items, events or data not conforming to expected patterns from dataset. In this paper, we apply evolution strategy to the anomaly detection in daily life monitoring. First, we explain how to use ISS for robot partners and wireless sensor networks. Next, we explain two main components of (1) human localization by spiking neurons and (2) daily life pattern extraction by Gaussian membership functions in the daily life monitoring. Next, we propose an anomaly detection method using evolution strategy. Finally, we present numerical experimental results and discuss the effectiveness of the proposed method.