Unobtrusive and Energy-Efficient Swimming Exercise Tracking Using On-Node Processing

Abstract

Body-worn sensors for movement analysis in swimming have to be unobtrusive and energy-efficient. We present a swimming exercise tracker for the unobtrusive positioning at the back of the head and an energy-efficient analysis using an on-node implementation. To develop the system, we collected head kinematics from 11 subjects in two 200-m medley races comprising breaks, turns, and four swimming styles. Each subject was equipped with a 6-D inertial measurement unit and completed one session in rested and fatigued state. Data were analyzed with a classification system, whereby different classifiers, window sizes, and feature sets were evaluated. Algorithm selection for on-node processing was performed on the basis of classifier accuracy and computational cost. The algorithm with the best tradeoff in accuracy and computational cost was selected and had a classification rate of 85.4%. Energy consumption of both on-node processing and Bluetooth streaming was evaluated on the Shimmer sensor platform. The results revealed energy savings of over 60% when data were processed on the sensor node. The presented analysis approach can be easily applied to other data analysis tasks, and the presented toolchain can support the rapid development of wearable systems in sports and healthcare.