Life pattern estimation of the elderly based on accumulated activity data and its application to anomaly detection

A life pattern estimation method and its application to anomaly detection of a single elderly are proposed. Our observation system deploys some pyroelectric sensors in an elderly's house and monitors and measures activities 24 hours a day to grasp residents' life patterns. Activity data is successively forwarded to the nurse operation center and displayed to nurses at the center. The system reports status related to anomalies together with the basic activities of elderly residents to the nurses, who decide whether recent accumulated data expresses an anomaly or not based on suggestions from the system. In the system, residents whose lifestyle features resemble each other are categorized into the same group. Anomalies that occurred in the past are shared in the group and utilized in an anomaly detection algorithm. This algorithm is based on an "anomaly score." The score is figured out by utilizing the activeness of the house's elderly resident. This activeness is approximately proportional to the frequency of sensor response within one minute. The anomaly score is calculated from the difference between activeness in the present and in the past averaged over the long term. The score is thus positive if activeness in the present is greater than the average in the past, and the score is negative if the value in the present is less than average. If the score exceeds a certain threshold, it means that an anomaly event has occurred. An activity estimation algorithm is also developed that estimates the basic activities of residents such as getting up in the morning, or going out. The estimation is also shown to nurses with the anomaly score of residents. Nurses can understand the condition of elderly residents' health by combining the information and planning the most appropriate way to respond.