The Alzimio App for Dementia, Autism Alzheimer's: Using Novel Activity Recognition Algorithms and Geofencing

摘要

Dementia, Autism, and Alzheimer's disorders affect millions of people worldwide. Suffering from forgetfulness, affected patients tend to wander off and potentially get into dangerous situations. This work introduces a mobile app, Alzimio, to provide two main alarm functions to these patients; safe-zone geofencing and activity-recognition. Caregivers designate certain activities (or zones) as dangerous, triggering alert message (with activity information) when detected. Several challenges must be overcome to achieve our goals. First, the activities and zones must be detected accurately in a timely fashion, and with high confidence. Second, the algorithms used should operate effectively on regular smartphones without special hardware. Such challenges are not unique to Alzimio, but are general to most Internet of Things (IoT) healthcare apps. In this study, we devise several novel activity-recognition algorithms to meet our goals of accuracy and efficiency. Our threshold-based algorithms intelligently filter and process the output of Android APIs for activity recognition and geofencing, at different time scales. The app was evaluated using extensive scenarios of usage for several months. We find that our max-in- window algorithm is able to achieve over 95% accuracy in less than 30 sec in most scenarios. The optimal threshold was found to be 65% confidence, to achieve best accuracy and delay. The Alzimio app runs efficiently on budget (low-end and medium) Android phones without noticeably affecting power consumption.