Anomaly Detection and Life Pattern Estimation for the Elderly Based on Categorization of Accumulated Data

Authors:
  • TPTuan D. Pham
    Tuan D. Pham
  • XZXiaobo Zhou
    Xiaobo Zhou
  • HTHiroshi Tanaka
    Hiroshi Tanaka
  • MOMayumi Oyama-Higa
    Mayumi Oyama-Higa
  • XJXiaoyi Jiang
    Xiaoyi Jiang
  • CSChangming Sun
    Changming Sun
  • JKJeanne Kowalski
    Jeanne Kowalski

Abstract
We propose a life pattern estimation method and an anomaly detection method for elderly people living alone. In our observation system for such people, we deploy some pyroelectric sensors into the house and measure the person’s activities all the time in order to grasp the person’s life pattern. The data are transferred successively to the operation center and displayed to the nurses in the center in a precise way. Then, the nurses decide whether the data is the anomaly or not. In the system, the people whose features in their life resemble each other are categorized as the same group. Anomalies occurred in the past are shared in the group and utilized in the anomaly detection algorithm. This algorithm is based on “anomaly score.” The “anomaly score” is figured out by utilizing the activeness of the person. This activeness is approximately proportional to the frequency of the sensor response in a minute. The “anomaly score” is calculated from the difference between the activeness in the present and the past one averaged in the long term. Thus, the score is positive if the activeness in the present is higher than the average in the past, and the score is negative if the value in the present is lower than the average. If the score exceeds a certain threshold, it means that an anomaly event occurs. Moreover, we developed an activity estimation algorithm. This algorithm estimates the residents’ basic activities such as uprising, outing, and so on. The estimation is shown to the nurses with the “anomaly score” of the residents. The nurses can understand the residents’ health conditions by combining these two information.