Body-worn accelerometer-based health assessment algorithms for independent living older adults

Abstract

The mainstream smart wearable products used for activity trackers have experienced significant growth recently. Among the older population, collecting long periods of activity data in a real-life setting is challenging even with wearable devices. Studies have found inconsistent and lower accuracies when older adults use these smart devices [1], [2],[2],[3]. As a person ages, many have lower daily levels of activity and their dynamic functional patterns, such as gaits and sit-to-stand transitional movements vary throughout the day. This thesis explores wearable health-tracking applications by evaluating daytime and nighttime pattern metrics calculated from continuous accelerometer signals. These signals were collected externally from the upper trunk of the body in an independent-living environment of 30 elderly volunteers. Our gold standard to validate the metrics from the accelerometer signals were similar metrics calculated from an in-home sensor network [4]. This thesis first developed an algorithm to count steps and another algorithm to detect stand-to-sit and sit-to-stand (STS) to demonstrate the importance of considering differences in daily functional health patterns when creating algorithms. Next, this thesis validates that accelerometer data can show similar motion density results as motion sensor data. And thirdly, this thesis proposes an updated vacancy algorithm using a new motion sensor system that detects when no one is in the living space, compared against the current algorithm.