Estimating running kinematics variability with an IMU sensor placed on the runner's thorax

Our contribution consists in a deep learning algorithm able to estimate running variability from a single IMU sensor placed on the runner's thorax. The algorithm has been trained and tested with data collected in indoor conditions, at different locomotion speeds and uphill/downhill grades. The reference values for running variability were provided by an opto-electronic system. Results suggest that running kinematic variables (stride frequency and duty factor) and variability (DFAa and Higuchi's D) could be estimated by the algorithm with an accuracy compatible to that obtained with a couple of inertial measurement units (IMU) placed at the foot level. Limitations apply, but it is suggested that the algorithm developed here could constitute a unique tool to assess running kinematic variables and variability indices with a single IMU placed at the thorax level.