Analysis of spatio-temporal parameters during running at different speeds using foot-mounted magneto-inertial sensors

Running characterisation can be highly valuable in predicting the onset of running-related injuries, thereby enabling their mitigation or prevention, and in assessing runner’s performances through the estimation of relevant spatio-temporal parameters, such as stride velocity and stride length. The existing literature presents a variety of approaches for the out-of-lab estimation of running spatio-temporal parameters utilising body-mounted magneto-inertial measuring units (MIMUs). The most widely used approach for the assessing of stride velocity and length is based on the double integration of feet accelerations between two consecutive ground contacts of the same foot. Specifically, the following steps need to be implemented. Firstly, foot orientation is estimated from the inertial data to remove the gravitational contribution from foot accelerations. Secondly, the zero-velocity update technique is implemented to detect the instants when the foot can be assumed to be stationary during the stance phase. The final step involves the removal of residual drift. As errors associated with each step propagate to the estimation of the final quantities, it is essential to identify sub-optimal values for each method composing the computational pipeline. However, this is not straightforward as the values of the optimal parameters may vary according to the running speed analysed. This thesis work dealt with the identification, implementation, and fine-tuning of state-of-the-art methods for the estimation of stride velocity and length under different running speeds. For this purpose, two datasets were analysed. The first dataset included shoe-mounted inertial data from 10 amateur runners running at 14 km/h on a treadmill, and the stereophotogrammetric data considered as a gold standard. For the second dataset 10 elite runners were enrolled to perform 50-m sprints at their maximal speed (19-27 km/h) and were instrumented with shoe-mounted MIMUs. Pressure-sensitive insoles and video recordings were used as portable gold standards. The results obtained from the MIMU-based pipeline were compared with the references provided by the available gold standards. The optimisation procedure enabled to considerably improve the results with respect to a standard non-optimised pipeline: the mean absolute percentage error on running stride velocity decreases from 17.8% to 10.5%, and from 16.5% to 11.1% for the first and second dataset, respectively. Thus, the proposed optimised pipeline is a promise solution for in-field running characterisation at different speeds.