Combining activities of daily living and scene understanding for continuous assessment of behavior patterns using depth data

Abstract

[ACCESS RESTRICTED TO THE UNIVERSITY OF MISSOURI AT AUTHOR'S REQUEST.] Research has shown the importance of monitoring behavior patterns of older adults over extended time periods to detect and deviations. These deviations can help determine health changes which could be prevented by early intervention. However, detecting these behavior patterns can be difficult using wearable sensors. These require the user to continuously wear the device which may not be acceptable for older adults. To address this, we propose an unobtrusive method to detect and measure the behavior patterns of older adults in their apartments using depth sensors. The many advantages of this approach include the low cost, unobtrusive, great performance in low illumination. An initial study, conducted in laboratory settings, detects activities in controlled settings. This study was further validated from data collected at TigerPlace, an independent facility for older adults in Columbia, MO. Sedentary behavior patterns of older adults with different physiologies and lifestyles are explored and results are reported. Activities of daily living (ADLs) are explored and a framework for recognizing ADLs in unstructured as well as unknown settings without prior information are explored. This can be coupled with activity information obtained from other sensors such as bed sensors, acoustic sensors, and motion sensors to generate more descriptive activity patterns. An in-home activity monitoring system would benefit greatly from our algorithm to alert healthcare providers of significant temporal changes in ADL behavior patterns of frail older adults for fall risk and cognitive impairment.