Normalization and extraction of interpretable metrics from raw accelerometry data

Abstract

We introduce an explicit set of metrics for human activity based on high-density acceleration recordings from a hip-worn tri-axial accelerometer. These metrics are based on two concepts: (i) Time Active, a measure of the length of time when activity is distinguishable from rest and (ii) AI, a measure of the relative amplitude of activity relative to rest. All measurements are normalized (have the same interpretation across subjects and days), easy to explain and implement, and reproducible across platforms and software implementations. Metrics were validated by visual inspection of results and quantitative in-lab replication studies, and by an association study with health outcomes.

Keywords: Activity intensity; Movelets; Movement; Signal processing; Time active; Tri-axial accelerometer.

Figure 1.

Two panels (from Subject 3208 and Subject 3056) of the raw three-axis accelerometer data with Active vs. Inactive prediction results. Each axis is illustrated in a different gray scale. The -axis stands for the time (in minute) from the first observation and the -axis shows values of the raw output from the accelerometer. The inactive time periods (according to our algorithm which is also described in this section) are shaded in light-gray.

Figure 2.

The density curves of standard deviations for all (34) subjects in different gray scale. The upper panel contains density curves of the original standard deviations and the lower panel contains density curves of standard deviations greater than . The vertical black lines in both panels are at .

Figure 3.

Two periods of raw data (panels 1 and 2), TA bars for Subject 3092 (panel 3). Each TA bar has a value between 0 and 1, and is colored light gray (), mid-gray (), or dark gray (). Panels 1 and 2 correspond to the light gray bars in the box frames depicted in panel 3. Panel 4 displays 3-day AI for Subjects 3029, 3056, and 3092. AI bars are colored light gray, mid-gray, or dark gray according to their TA values as in panel 3.

Figure 4.

Top panels: CRTA for Subject 3029 (a), 3056 (b), and 3092 (c). Bottom panels: CRAI for Subjects 3029 (d), 3056 (e), and 3092 (f). There are three curves in each plot, each representing 1 day. These curves are either compressed or stretched so that they display CRTA or CRAI in the scale of percentage time of the day, instead of actual time.

Figure 5.

The plots for the metrics validation for Subject 3056 and Subject 3092 during two different visits. In each visit, there are two replicates of walking and three replicates of chair-stands. The raw data during these periods as well as the AI were plotted below the probability density curves of AI. In the plots of raw signals and plots of AI, each gray grid stands for 1 s. A lot of similarities of AI can be observed within and across subjects, while the modes of the distributions are very close. Also, AI picked up the change of variability of the signal during sudden movements such as standing-up-from-chair and assigned low values to the resting periods (both standing and sitting).