Activity Recognition System for Dementia in Smart Homes based on Wearable Sensor Data
Chun-Fang Su, Li‐Chen Fu, Yi-Wei Chien, Ting-Ying Li · 2018
The highly developed medical technology has increased the average human life span, but also speeds up the process of the world's population aging. In the next few decades, Dementia will affect a considerably increasing number of the elderly. Taking care of the person with dementia is an overwhelming task and can cause physical, emotional and economic pressures to families and caregivers. This works aims to alleviate the caregiver's burden, which can constantly monitor the elderly in smart homes, and in turn, send alert to the caregiver whenever abnormal activities of the elderly occur. This system adopts the unsupervised clustering algorithm, e.g. Dirichlet process mixture model (DPMM), to get the potential clusters about the hands movements because the number of clusters is unknown in priori. In order to store spatial information of hands movements within a special activity, N-gram model is used to extract the feature basing on the result of DPMM, and in turn integrate position and hands movement features. The activity recognition (AR) model is used to recognize the living activities with promising performance, in which the precision and recall of this proposed system is up to 96.8% and 96.7%, respectively, even for the similar activities, such as “eating mear' and “having medicine.