Running analysis on treadmill and on track based on magneto-inertial sensing technology

Running temporal parameters, i.e., running cycle, stance and swing durations, are an effective way to evaluate running performances. Running characterization can be useful to predict the advent of injuries, enabling to reduce their entity or prevent them. During the last decades, magneto inertial measurements units (MIMUs) have become the most widespread wearable solution to investigate running in outdoor conditions, evaluating the runner’s actual performances. In the literature there is an extensive number of different methods for the detection of temporal events; however, each method is usually targeted to a restrict range of running speeds due to the high variability of the morphology of the inertial signals varying running paces. The aim of the present work is to perform a comparative evaluation of different state of the art methods for the estimation of running temporal events (i.e., instants of initial, IC, and final contacts, FC, with the ground) across different running paces. Nine methods selected from the literature were implemented and adapted to the collected data based on the different sampling frequencies and sensor locations. In addition, an original template-based method, using Wavelet Transformations and Dynamic Time Warping suitable for accurately defining and segmenting the running cycle on a wide speed range (8-32 km/h) was implemented. For the aim of this thesis, three datasets were analysed. All the recruited subjects were equipped with a MIMU fixed to the shoelaces of each shoe. The first dataset included 11 subjects that ran at two different constant speeds (8 km/h and 10 km/h) in outdoor and indoor conditions, equipped with MIMUs and sensorized pressure insoles, considered as a portable gold standard, both sampled at 100 Hz. The second dataset included data of 10 amateur runners who were asked to run at 14 km/h on a treadmill, instrumented with MIMUs sampled at 200 Hz and retro-reflective markers, since the stereophotogrammetric system was taken as the gold standard. Lastly, the third dataset included 9 elite runners, whose speeds ranged from 20 to 32 km/h and was on an outdoor running track, adopting sensorized pressure insoles as gold standard, acquired at 100 Hz. Comparing the performances on the three datasets in terms of root mean square errors (RMSE) of running events and mean absolute percentage errors (MAPE) of running phases against the available gold standard, Blauberger et al. (2021) was selected as the best tradeoff for the temporal parameters estimation, resulting in a RMSE lower than 0.04 s on IC, and lower than 0.06 s on FC for all the speeds analysed. Furthermore, the results obtained via the novel method were compared to the performance obtained with the one proposed by Blauberger et al. The proposed method reported a RMSE lower than 0.033 s on the detection of ICs, and an RMSE lower than 0.049 s on the detection of FCs for all the different running paces. These results were deemed comparable to the ones achieved with the most suitable method from the literature for different paces, thus the proposed method is promising for the detection of temporal parameters on a wide speed range and in out-of-lab applications.