Improving Complex Nurse Care Activity Recognition Using Barometric Pressure Sensors

Abstract

In this paper, we investigate complex nurse care activity recognition using barometric pressure sensors and show several characteristics of pressure features, such as identifying activity classes which can improve when we use the barometric pressure sensors and investigating the relationship of pressure features with the modification of label durations which are often required in complex and realistic applications. For modification of label durations, we propose timestamp extension methods using sequential and dynamic with two approaches (T1 and T2) for inaccurately labelled behaviors. We also consider how the speed of movement in activities affects the barometric pressure sensor readings aside from posture or height. Using a barometric pressure sensor, the accuracy of activity recognition can improve as much as 24 activities from 33 activity classes, even the activity of posture changes increase by 15% because the barometric pressure sensor not only affects large movements but also subtle movements of the human body. This makes the pressure sensor become important in recognizing complex activities, particularly in nursing care. We collect real-world data in hospitals, therefore we also present a system that is effective in gathering data for activity recognition without interfering with the nursing tasks. By using the systems, we gathered 4,544 activity labels from 15 participants.