| Autor: |
Singh, Pratham, Esposito, Michael J. S., Barrons, Zach B., Clermont, Christian A., Wannop, John W., Stefanyshyn, Darren J. |
| Předmět: |
|
| Zdroj: |
Sports Engineering (Springer Science & Business Media B.V.); 8/24/2022, Vol. 25 Issue 1, p1-9, 9p |
| Abstrakt: |
Machine learning methods such as a stepwise linear regression and a feedforward neural network were implemented and analyzed for their capability of estimating stride length using wearable sensors. Strides were segmented using acceleration data from a sacrum mounted inertial measurement unit. Stride length was then obtained by calculating the difference in position from a local positioning system (LPS) operating in the ultrawide bandwidth. Using the last stride of the trial, 27 predictor variables from the LPS were analyzed. Position data from motion capture (MOCAP) were used (i.e. the reference system) for determining accuracy of each machine learning method. Variables from the stepwise linear regression model were used as input nodes for the neural network. To avoid overfitting the data in the neural network, a gradient descent optimization algorithm was used. Based on an a priori analysis, 90 participants were recruited and performed three self-selected walk, run and sprint speeds. Upon analyzing, the lowest normalized root mean square error was 0.08 by the feedforward neural network (i.e. the comparator) after receiving inputs from the regression model and learning stride length estimates from MOCAP. Therefore, the neural network accurately estimated stride length when learning from a reference system such as motion capture. [ABSTRACT FROM AUTHOR] |
| Databáze: |
Complementary Index |
| Externí odkaz: |
|
Full text from SpringerLink