Title: Walking recognition via continuous wavelet transforms applied to the longitudinal intervention study
Authors: Marcin Straczkiewicz - School of Public Health-Bloomington, Indiana University (United States) [presenting]
Christopher Sorensen - School of Medicine Washington University in Saint Louis (United States)
Jacek Urbanek - Johns Hopkins Bloomberg School of Public Health (United States)
Beau Ances - School of Medicine Washington University in Saint Louis (United States)
Ciprian Crainiceanu - Johns Hopkins Bloomberg School of Public Health (United States)
Jaroslaw Harezlak - Indiana University School of Public Health-Bloomington (United States)
Abstract: Recent advancements in accelerometry provide researchers with more objective ways to quantify physical activity (PA) than the commonly used questionnaires. Wearable devices enable the quantification of the behavior in a free-living environment. However, the common summary measures, e.g. activity counts, lose crucial information from the signal. We developed an algorithm (RoWW - Recognition of Walking using Wavelets) based on the Continuous Wavelet Transform (CWT) to classify the walking vs. non-walking periods using raw accelerometry data. Our algorithm quantifies the characteristics of the walking periods including its periodicity, duration, speed and intensity. Our CWT-based algorithm is sensitive to subtle changes in periodic content of a signal and detection of its breakdown points or signal discontinuities. RoWW automatically fragments the time-frequency display of a given signal and identifies the time segments with the pronounced periodicity. We apply RoWW to study the influence of two types of behavioral interventions on walking habits among a cohort of HIV-infected patients, whose data were collected for one week at baseline and at a 3-month visit.