Using Barycenters as Aggregate Representations of Repetition-Based Time-Series Exercise Data
Bahavathy Kathirgamanathan, James Davenport, Brian Michael Caulfield, Pádraig Cunningham · Communications in computer and information science · 2022
This paper introduces the use of time-series barycenter averaging as a means of providing aggregate representations of repetition-based exercises. Time-series averaging is not straightforward as small misalignments can cause key features to be lost. Our evaluation focuses on the Forward Lunge exercise, an exercise that is used for strengthening, screening and rehabilitation. The forward lunge is a repetition-based movement so assessment entails comparing multiple repetitions across sessions. We show that time-series barycenters produced using Dynamic Time Warping are effective for this application. The barycenters preserve the key features in the component time-series and are effective as an aggregate representation for further analysis.